feat: add client-side audio chunking for large file transcription

[NathanSkene]

Summary

• Splits large audio recordings into smaller chunks for transcription
• Enables transcription of longer recordings without API timeout issues
• Chunks are processed sequentially and results concatenated

Test plan

• Record a long dictation (>30 seconds)
• Verify transcription completes without timeout
• Verify output text is coherent across chunk boundaries

[NathanSkene]

Manual validation update on macOS with the packaged installed app:

This feature passed validation.

What I tested:

• a genuinely long BYOK transcription using meeting audio
• the transcription completed successfully instead of failing or timing out
• the output stayed coherent across the full recording without obvious chunk-boundary corruption (no abrupt truncation, duplicated joins, or missing blocks)

Follow-up note: the output still needs better formatting / paragraphing for long meeting transcripts, but that is a separate UX/output-quality issue rather than a failure of the chunking feature itself.