feat(scoring): surface the merged-PR history floor in the score breakdown - #1801

…wn (#1877)

explainScoreBreakdown explained every other scoring multiplier (density,
contribution bonus, label, issue, credibility, review penalty, open-PR
pressure, open-issue spam, merged-PR history floor in #1801, and the
issue-discovery validity floor now in #1874) but silently omitted the
timeDecayMultiplier, even though it multiplies into the final score
(preview.ts:432) and is the only remaining multiplier on scoreEstimate
without a breakdown component.

Add a timeDecayBreakdown: neutral when the multiplier is >= 0.99 (the
PR is fresh or upstream time-decay -- env SCORING_TIME_DECAY_ENABLED,
default OFF -- is disabled for the preview), and reduced with an aged-PR
lever when the multiplier has decayed. Purely additive explanation; no
scoring behavior change.

…eakdown (#2356)

* feat(scoring): surface the saturated base-score value in the score breakdown

explainScoreBreakdown explained every multiplier on scoreEstimate that
scales the score (density, contribution bonus, label, issue, credibility,
review-penalty, review-collateral, open-PR pressure, open-issue spam, the
merged-PR history floor from #1801, the issue-discovery validity floor from
#1984, and the time-decay floor from #1877) but silently omitted the
**baseScore** itself — the saturated-value foundation
`25 × (1 - exp(-src_tok / SRC_TOK_SATURATION_SCALE=58))
 + min(total_token_score / CONTRIBUTION_SCORE_FOR_FULL_BONUS=1500, 1) × MAX_CONTRIBUTION_BONUS=5`
(capped at 30) that flows into estimatedMergedScore before any multiplier
applies. A contributor could see their `densityMultiplier` was healthy but
had no surface for the actual cap contribution or the contribution-bonus
adder.

Add a `baseScoreBreakdown` sibling of `densityBreakdown` (which already
uses `baseTokenGatePassed` as a gate trigger) that:

- returns `blocked` when `baseTokenGatePassed` is false (the change does
  not yet meet the minimum meaningful source-change threshold),
- returns `full` when baseScore is saturated near the 30-point cap and
  names whether the contribution bonus is contributing,
- returns `neutral` when baseScore is mid-curve, with a copy that names
  BOTH the baseScore and the contributionBonus dimensions explicitly
  (avoids the §7 Tier A rule 3 collapse-copy nit — the production copy
  always iterates on observed values for both axes, no "no source
  contribution observed" single-clause ever produced).

Purely additive explanation projection; no scoring behavior change.

Wiring:

- src/services/score-breakdown.ts: add `baseScoreBreakdown`, insert
  at index 0 of the components projection (foundation before density /
  contributionBonus), under the `leverageScore` convention that keeps
  `densityMultiplier` as the canonical top lever when both are blocked
  (baseScore blocked = 70, below densityMultiplier's 75).
- test/unit/score-breakdown.test.ts: include the new `baseScore`
  component in the top-level `arrayContaining` regression list, and add
  a focused test covering the blocked / neutral-with-bonus / saturated
  branches (avoids the §7 Tier A rule 5 missing-both-branches test gap).

Verified locally:

- `git diff --check` clean.
- `npm run typecheck` clean.
- `npm run actionlint` clean.
- `npm run db:migrations:check` — 90 migrations OK, contiguous 0001..0087.
- `npm run build:mcp` clean.
- `npx vitest run test/unit/score-breakdown.test.ts --coverage
  --coverage.include='src/services/score-breakdown.ts'` —
  **15/15 tests pass**; 100% statements; 100% lines; 3 uncovered
  branches in pre-existing code (issueMultiplierBreakdown:240 +
  2 in pre-existing reviewCollateralBreakdown), not touched by this change.

Diff stat: `src/services/score-breakdown.ts` +34 lines,
`test/unit/score-breakdown.test.ts` +60 lines, total +94 lines
(`size:S` territory; orb counts src files primarily per §7 Tier A rule 1).

* fix(scoring): address review nits on base-score breakdown

- Fix test fixture MAX_CONTRIBUTION_BONUS from 25 to 5 (match production)
- Replace hardcoded '30-point cap' with generic 'score cap' text
- Extract BASE_SCORE_SATURATION_DISPLAY_THRESHOLD (29.5) as named constant
- Bump baseScore blocked leverageScore from 70 to 75 (match density gate block)
- Split 3-branch test into separate focused it() blocks
- Add test for hasBonus=false + gate-passed branch

* fix(scoring): derive base-score saturation from preview-estimate cap, not hardcoded threshold

Hardcoding BASE_SCORE_SATURATION_DISPLAY_THRESHOLD=29.5 meant any preview
using different MERGED_PR_BASE_SCORE or MAX_CONTRIBUTION_BONUS constants
would mislabel saturated vs sub-cap.

- Add baseScoreCap to ScorePreviewResult.scoreEstimate (computed alongside
  baseScore in the scoring core, which has the snapshot constants)
- Replace hardcoded 29.5 with BASE_SCORE_SATURATION_RATIO=0.95 applied to
  baseScore / baseScoreCap — always relative to the active model's cap
- baseScoreCap is undefined when fixedBaseScore is in effect
- Add regression test with MERGED_PR_BASE_SCORE=50, MAX_CONTRIBUTION_BONUS=10
  proving the threshold adapts to non-default constants

* fix: guard baseScoreCap > 0 before division, fix test comment

- Add baseScoreCap > 0 guard before saturation ratio division
- Fix test comment: 993 is the 95% threshold value, not 'full bonus'

* fix: remove dead locals from test, handle baseScoreCap undefined copy

* test: cover fixedBaseScore and saturation model cap branches

- fixedBaseScore override → baseScoreCap undefined, copy says 'fixed base score override'
- pending_saturation_model → cap = MERGED_PR_BASE_SCORE + MAX_CONTRIBUTION_BONUS