fix(player): stabilize multi-audio switching

[stackia]

Summary

• Reuse the PCM audio player, AudioContext, and fixed output graph across same-slot stream reloads while fully resetting stream-specific state.
• Make WSOLA initialization generation-safe so stale async work cannot attach to a new stream.
• Stabilize internal and alternate audio switching across AAC, MP2, and unsupported AC-3/E-AC-3 tracks.
• Keep unsupported tracks selected, show the existing audio warning, and preserve video playback with silent AAC timeline occupancy.
• Reset PCM timing and startup synchronization consistently when returning from unsupported or MSE-decoded audio.

Root cause

Audio switches mixed stream-level state with long-lived Web Audio resources. Stale PCM scheduling and WSOLA initialization could survive a reset, while unsupported AC-3 caused the MSE audio transition to fail and roll the worker back to the previous track. Subsequent switches then inherited inconsistent timing and decoder state, resulting in silence or intermittent audio.

User impact

Same-slot channel changes now reuse the existing Web Audio runtime without leaking audio from the previous stream. AAC and MP2 switches recover reliably. Selecting unsupported AC-3/E-AC-3 remains on that track, displays a warning, and continues silent video playback until the user selects a supported track.

Validation

• pnpm exec biome check ...
• pnpm run type-check:tsc
• pnpm exec vite build web-ui
• Release C build
• uv run ruff check e2e
• uv run pytest e2e --collect-only -q (539 tests)
• ./scripts/run-e2e.sh --co (539 tests)
• Chromium dev lab: HTTP TS internal, HLS-TS internal, and HLS-TS alternate audio switching across AAC, MP2, and AC-3
• Volume, mute, pause/resume, startup sync, and repeated same-slot switching

[github-actions]

Documentation preview

The documentation preview has been deployed for this pull request.

• Preview: https://thankful-water-0a297bf00-677.eastasia.1.azurestaticapps.net
• Commit: e85ab75

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: abc1b0a6d4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Handle timestamp restarts in silent audio-only tracks

When the selected track is an audio-only TS pipeline using silent AAC occupancy (for example an alternate AC-3/E-AC-3 rendition on a browser where that codec is unsupported), these raw firstDts - _dtsBase values are passed to _generateSilentAudio, which keeps _silentAudioLastDts as a monotonic output cursor. If the audio rendition crosses a TS timestamp restart and end moves behind the previous cursor, _generateSilentAudio emits no samples until the source timestamps catch up, leaving the audio SourceBuffer without the promised silent occupancy and risking a stall instead of continuing silent video playback.

Useful? React with 👍 / 👎.