Fix Emby HEVC 10-bit SDR quality-reopen corruption

[jlipworth]

Summary

• Fix Emby HEVC 10-bit SDR: block/smear corruption after Original to 8 Mbps #304's reproduced HEVC 10-bit SDR quality-reopen corruption with full-timeline native resume/seek, preserving negotiated session/audio/container and bitrate ceilings.
• Rebased onto the landed Fix Jellyfin HEVC HLS packaging and add private corpus testing workflow #306/Fix Emby AV1 cold-start deadline teardown cycle #307 changes; dropped duplicate tooling/identity commits and preserved AV1 warm-up before the disjoint HEVC lane.
• Condense corpus, AV1 and HEVC Markdown; retain current policy in playback architecture and unresolved evidence/gates in research notes.

Validation

• 1,704 hermetic PMSKit tests; hygiene, strict MkDocs, links/anchors and Mermaid checks.
• Four fresh native builds and owned serial platform smokes; installed simulator binaries match products.
• Post-rebase exact-source Original → 8 Mbps preserved progress with 4+4 inspected clean HEVC frames; native seek back to 10 minutes also passed with 4+4 clean frames.
• HEVC 8-bit and H.264 transition controls passed without selecting the HEVC lane. H.264 post-transition frames include a dark title fade, not a broad scene-quality claim.
• AV1 integration regression warmed in 6.539s and passed zero-seek/hold with 4+4 inspected visible frames. No remaining server encoder after stop.

Limits

The reproduced boundary is legacy offset priming/proxy startup, not a proven fragment/init/decoder mechanism or a shared cause with #291. Physical-device, broader server-version and full lifecycle/seek acceptance remain open. HDR is excluded. Known nonfatal simulator audio-session/GroupActivities warnings remain outside this fix.

Private source identifiers, names, addresses, paths, credentials and frames are not attached. No automatic issue closure or merge.