Regenerate transcript for a call from the saved audio recording.

This endpoint allows regenerating the transcript when the original transcription had issues (e.g., speaker misidentification).

Requirements:

• The call must have a saved audio recording in S3
• User must have access to the call

The regeneration process:

For normal calls (not transferred):

1. Resets the transcript status to pending
2. Resubmits the audio to AssemblyAI with 1-2 expected speakers
3. Uses LLM-based speaker identification for better accuracy
4. Re-runs grading and summarization after transcription completes

For transferred calls:

1. Resets the transcript status to pending
2. Downloads the audio and splits it at the transfer point
3. Transcribes bot segment (Caller + Assistant) separately
4. Transcribes attorney segment (Caller + Attorney) separately
5. Merges transcripts with correct speaker labels
6. Re-runs grading and summarization after transcription completes

Returns 202 Accepted since the transcription happens asynchronously.