Add audio playback to sample session fixtures

[btc]

Context

The sample session fixtures (internal/sample/fixtures/) are extracted from v0 session 27 ("Chat System"). Currently all audio_url fields are null — the fixtures are text-only. This means the sample session page at /sample has no audio playback capability.

Goal

Add TTS and STT audio to the sample fixtures so the public demo can showcase the full interview experience, including session replay with audio.

What's needed

• STT audio (candidate): The original candidate audio recordings from session 27 should be on disk from the v0 prototype. Locate and include them.
• TTS audio (interviewer): The AI interviewer responses need TTS audio generated. This may require re-running the TTS pipeline against the session 27 transcript to produce audio files for the interviewer turns.
• Storage: Sample audio files need to be served publicly without auth. Options: embed in the binary (if small enough), serve from object storage with public URLs, or a dedicated static asset path.
• Fixture updates: Update session.json message entries with actual audio_url values pointing to the served audio.
• Replay integration: The session replay engine (web/src/components/replay/engine.ts) and transcript page already support the replay timeline — audio playback needs to sync with message reveal during replay.

Notes

• Current sample fixture provenance: v0 prototype, session 27
• The extract-sample script may need updating to also export audio files
• Consider file size — embedding large audio in the binary may not be practical