chore: sync fork with upstream (v2.5.6, #338–#456) - #83

…after Back (chattymin#435)

* test(collection): add a pure page lookup for the Pokédex grid

DexScrollPager.page(containing:in:pageSize:) returns the page that shows
a species, or 0 when it is nil or filtered out. Not wired up yet; the
next commit uses it to return to the viewed page after Back.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* fix(collection): return to the viewed Pokédex page after Back

The detail page replaces the grid, so Back rebuilt DexGridView and its
page state started over at page 1. CollectionView now remembers the
species whose detail page was closed and the rebuilt grid starts with it
selected, deriving the page from that selection. Opening a species from
a Home sprite now also lands on its page on Back. The anchor is used
once, so a later rebuild (segment switch) still starts on page 1.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* fix(collection): keep the catch log row in view after Back

The catch log sits in the same branch the detail page replaces, so Back
scrolled it to the top. The row that opened the detail page is now
scrolled back to the center. The scroll waits one turn after the log
appears: before the first layout the lazy rows above it are estimated
and the scroll stopped short of the row.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* docs: log the detail-page remount that reset the Pokédex page

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* docs: show Pokédex and catch log Back comparison

Sample-data renders of the real CollectionView for the PR.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* fix(collection): correct the catch log scroll after rows are measured

CI showed the restored log stopping a row short of the row that opened
the detail page. The first scrollTo places the target with estimated
heights for the lazy rows; once the rows it reveals are measured they
push the target off its spot. Scroll again after that layout pass. A
single scroll fails the test locally too.

The tests also retry a press until the screen changes: right after a
transition the view tree can still hit-test the previous screen, which
made a Back press miss on CI.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* docs: record the catch log scroll CI finding

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* fix(collection): set the grid return anchor only when Back lands on the grid

Back always stored the closed species as the grid's return anchor, but
only the grid's onAppear clears it. When Back returned to the catch log
(the detail page was opened from a log row, or from a Home sprite while
the log segment was selected), the anchor stayed, and the next switch
to the Pokédex segment opened on that species' page with it selected
instead of page 1.

testBackToTheLogLeavesTheGridOnItsFirstPage covers that path and fails
when the anchor is stored unconditionally again.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* fix(collection): reopen the Pokédex page after Back without selecting the species

The grid found its return page by starting with the closed species as
its selection. That brought back the selection border and the footer
line, which nothing had shown since the detail page started replacing
the grid (chattymin#264), and made the grid behave unlike the catch log, which
only scrolls back to the row.

DexGridView now keeps the anchor in its own state and uses it only to
pick the page. testBackFromDetailReturnsToThePageOfThatSpecies asserts
that the footer does not name the species and fails when the anchor
selects it again.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* docs: re-render the Pokédex Back comparison without the selection

Sample-data render of the real CollectionView after Back: page 2 with no
species selected.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Generalize the Unown letter forms (chattymin#288) to every generation 1-5 species
whose forms all have an animated Black/White sprite: Castform, Deoxys,
Burmy/Wormadam, Cherrim, Shellos/Gastrodon, Rotom, Giratina, Shaymin,
Arceus, Basculin, Darmanitan, Deerling/Sawsbuck, the forces of nature,
Kyurem, Keldeo, Meloetta and Genesect.

The form belongs to the individual, is drawn once for the whole line and
is resolved per species. Saves use form/pendingForm/representativeForm and
still read the former Unown keys.

…n#446)

* fix(settings): collapse automatic backup history by default

* test(settings): reveal snapshot disclosure before locating it

* test(settings): stabilize native snapshot capture and disclosure clicks

* fix(settings): use a native button for backup history

* test(settings): capture snapshot history across native viewports

* test(settings): recognize native snapshot renders in viewport strips

Correct Sonnet 5.5 cache-read estimates across persisted usage caches. Carry the current form into backup summaries and thumbnails. Preserve Antigravity refresh credentials after transient grant failures. Explain default-form metadata beside alternate appearances, with native rendering and focused regression guards.

…min#448)

Retain SwiftUI App startup and replace the unused Settings { EmptyView() } scene with _EmptyScene(). Settings remain in the existing popover; startup, menus, activation policy, and popover placement otherwise match main.

Stabilize the Korean chart OCR fixture using exact captions among equally highest-confidence candidates. Preserve diagnostic candidates and CI fixture images.

Validation: final PR build passed; 1,520 local tests excluding window-presenting suites passed with 16 skipped and zero failures. The empty-scene entry point compiles against SDK14.5 and26.5 targeting macOS14. Full CI passed: 1,560 tests, 16 skipped, zero failures, 93.80% logic-core line coverage; secret scan passed. Local app replacement is deferred to avoid desktop interruption.

…ymin#455)

…ole files (chattymin#338)

On an account with a large Claude Code history (~2GB of session logs in
the current month, single files up to 274MB) the cold scan pegged a core
for over a minute without finishing, and because the blob cache is keyed
on whole-file mtime+size, the active session file was re-parsed from byte
zero on every refresh.

Three changes:
- ISO8601Parser reuses one lock-guarded formatter pair instead of
  allocating an ISO8601DateFormatter per line (the dominant sampled cost).
- parseClaudeFile scans lines at the byte level via Data/FileHandle rather
  than materializing the file as a String and splitting into Substrings.
- Blob records a per-file byte offset; when a file has only grown, only
  the appended tail is parsed and merged. Same-size or shrunk files are
  re-read from the start. Older cache files without the key still load.

Measured on the history above: cold start ~15s then idle (was 60s+ and
never completed), steady-state refresh stays at 0% CPU.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Patrick.Park <pdm001125@naver.com>

* fix: parse Claude session logs incrementally instead of re-reading whole files

On an account with a large Claude Code history (~2GB of session logs in
the current month, single files up to 274MB) the cold scan pegged a core
for over a minute without finishing, and because the blob cache is keyed
on whole-file mtime+size, the active session file was re-parsed from byte
zero on every refresh.

Three changes:
- ISO8601Parser reuses one lock-guarded formatter pair instead of
  allocating an ISO8601DateFormatter per line (the dominant sampled cost).
- parseClaudeFile scans lines at the byte level via Data/FileHandle rather
  than materializing the file as a String and splitting into Substrings.
- Blob records a per-file byte offset; when a file has only grown, only
  the appended tail is parsed and merged. Same-size or shrunk files are
  re-read from the start. Older cache files without the key still load.

Measured on the history above: cold start ~15s then idle (was 60s+ and
never completed), steady-state refresh stays at 0% CPU.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(home): label the usage recap control

The chart icon was the only way into week, month, and year recaps, and it did not look like a button. Show a short localized label inside a light capsule.

* fix(home): preserve period values beside the recap label

* test(home): match OCR languages to period fixtures

---------

Co-authored-by: notcadenwhitt-stack <notcadenwhitt@gmail.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Patrick.Park <pdm001125@naver.com>

Use an unambiguous word for the oldest catch while retaining exact OCR,
click, Back navigation and scroll reset assertions. Adapt the fixture
repair from c7c2d7a and record the CI failure in the defect log.

Co-authored-by: 서민수 <sms@bbros.kr>
Co-authored-by: Codex <noreply@openai.com>

Merge chattymin/PokeTokenBar main (17 commits) into the fork.

Conflict resolutions that adapt upstream changes to fork-only code:
- Sonnet 5.5 cache-read price and the locked ISO8601 parser ported into PokeTokenBarShared,
  where the fork keeps ModelPricing and ISO8601Parser; the Mac keeps its typealiases.
- Empty Settings windows (chattymin#448): adopted _EmptyScene(). Without a Settings scene there is no
  system ⌘, item, so SettingsMenuItem.install adds the fork's own app-menu item (retargets an
  existing one if present). The unreachable SettingsSceneBridge fallback is removed.
- Antigravity transient refresh failure: ported upstream's fix onto the fork's
  AntigravityRefreshOutcome model. An expired Keychain credential whose grant fails temporarily
  now throws AntigravityTokenRefreshError.transient instead of returning the dead token (which
  401'd and invalidated the only refresh token). Upstream's tests rewritten onto tokenRefresher.
- Generic Pokémon forms (chattymin#376): the iPhone payload stays Unown-only; the export maps the new
  PokemonForm API onto it and no longer risks sending other species' forms as Unown letters.
- README, SettingsView state and defect-log: kept both sides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>