Harden experimental P7 fallback initialization identity

[jlipworth]

Summary

• Pin the complete validated initialization representation per proxy open, rejecting changed full/range responses rather than mixing bytes from distinct representations at one URL.
• Make replacement-wait regression coverage deterministic without extending deadlines or weakening identity, consent, cancellation, or progress checks. Live clock/polling defaults remain unchanged.
• Include isolated test-only prerequisite f56afe8 (also on codex/season-deletion-test-barrier): await deletion persistence before reopening the integration-test index. No download production behavior changes.
• Record exact-revision validation and limitations in the active DV research note.

Scope

The P7 HDR10-base candidate remains default-off, DEBUG macOS only. This is not native Dolby Vision P7 reconstruction or production enablement. Plex Generic, P5 guards, explicit video-encoding consent, source/session authority, and Release behavior remain unchanged. Related investigation: #316 (no closure claim).

Validation

• PMSKit: 1,707 Swift Testing + 121 XCTest cases pass, including initialization identity/range rejection and fresh-open coverage.
• Integrated full macOS hosted suite: 712/712 pass in three consecutive runs.
• iOS: 696/696 pass with ancillary simulator diagnostic collection disabled; default diagnostic collection hung after test completion and remains a runner limitation, not a green default run.
• tvOS: 383/383 pass; clean iPhone/tvOS semantic fixtures have inspected attachments.
• Fresh generic visionOS/iOS/tvOS builds pass; docs, links, Mermaid and hygiene checks pass.
• Exact-source native Plex P7 playback/seek/reopen and audio transitions have inspected evidence at the documented revisions. Fresh post-test-fix Plex seek/audio checks pass with video copy and inspected PQ/BT.2020 frames.
• Latest same-identity permission recheck: prior local-network denial no longer reproduces. Emby audio transition passes with inspected initial/post frames. Jellyfin authenticates and plays but fails the unchanged progress oracle after a 2.135-second observation gap; no post-hold pass. Plex's saved exact-source metadata binding returns HTTP 404, so that fresh reopen is blocked, not passed.

Remaining gates / non-goals

• Rebind the exact Plex source to finish the latest reopen validation; do not substitute a different source silently.
• Jellyfin sampling-gap/probe reliability remains a recorded limitation, not an attributed product regression or a waived test.
• visionOS run verification remains blocked by the missing canonical golden-simulator pointer.
• Broader source coverage, later-segment extracted-base equivalence, physical HDR/DV/audio acceptance, and production enablement remain separate.
• Prior failures and blocked runs remain documented; later passes do not retroactively erase them. Private media identifiers, credentials, frames, and raw logs are excluded.

No video encoding was authorized, and no release/upload is requested.