Add whisper.cpp transcription provider

[jamilahmadzai]

Summary

• Add a local --api whispercpp transcription provider backed by the whisper.cpp CLI.
• Convert input video/audio to 16 kHz mono WAV before invoking the local binary, matching whisper.cpp's documented CLI workflow.
• Document WHISPERCPP_BINARY, WHISPERCPP_MODEL_PATH, WHISPERCPP_THREADS, and WHISPERCPP_EXTRA_ARGS in the README and .env.example.
• Add focused unittest coverage for configuration validation, CLI command construction, output cleanup, file processing, and sapat --api whispercpp routing.

Why

This gives Sapat a no-cloud transcription path for users who want to keep recordings and transcripts inside a local or Daytona workspace. It does not require API keys, does not upload audio to a provider, and keeps the existing OpenAI/Groq/Azure paths unchanged.

This is the companion implementation for the Daytona content bounty submission: daytonaio/content#246

Validation

• .venv/bin/python -m unittest discover -s tests -v
• .venv/bin/python -m compileall src tests
• .venv/bin/sapat --help
• .venv/bin/python -m pip check
• git diff --check

No model files, private audio, secrets, or generated transcripts are committed.

Payment

Bounty claim and payout details are handled in the Daytona content PR: daytonaio/content#246

[jamilahmadzai]

Status note after the force-push with signoff:

The three Continuous AI checks currently show Agent encountered an error, and the target URL responds with Cannot GET /api/agent-check-summary/..., so I do not have actionable feedback from those bots. Local validation is still green:

• .venv/bin/python -m unittest discover -s tests -v
• .venv/bin/python -m compileall src tests
• .venv/bin/sapat --help
• .venv/bin/python -m pip check
• git diff --check

No secrets, private audio, generated transcripts, or model files are included.