Add Hugging Face ASR provider

[daxia778]

Summary

• Add a Hugging Face Inference Providers ASR backend exposed as --provider huggingface.
• Use HF_TOKEN, optional router provider/base/timeout env vars, and the current router model endpoint for binary audio transcription.
• Parse transcript text plus optional timestamp chunks into TranscriptionResult.
• Cover auth, endpoint construction, base64 payloads, timestamp/temperature parameters, availability, API errors, and malformed responses with mocked tests.

Validation

• .venv/bin/pytest (182 passed, 1 LibreSSL urllib3 warning)
• .venv/bin/black --check sapat/providers/huggingface.py tests/providers/test_group_a.py
• .venv/bin/python -m compileall sapat
• git diff --check

This is the companion implementation for a Daytona content guide attempt on daytonaio/content#13. No API keys, private media, generated transcripts, or payout details are included.