Auto-timeout long-running transcriptions

[julien731]

Story

As a user,
I want transcriptions that run too long to automatically fail with a clear message,
so that I'm not left wondering if it's still working.

Acceptance Criteria

• TRANSCRIPTION_TIMEOUT env var (default: 1800 seconds) added to config.py
• Transcription thread checks elapsed time at each stage transition
• If timeout exceeded, job marked FAILED, meeting marked ERROR
• Error message: "Transcription timed out after {minutes} minutes"
• Timeout value logged at warning level

Technical Notes

• Check timeout in _run_transcription before each major stage (transcribe, align, diarize)
• Add started_at timestamp to job tracking or use local variable in the transcription function

Size

M (4-6 hours)

Spec

docs/specs/transcription-failure-recovery.md — F3

Dependencies

Depends on #10 (error field in MeetingMetadata).