feat(stt): multi-provider STT with TranscriptionProvider trait

[rareba]

Supersedes #2995 (branch prefix correction: feature/ -> feat/)

Summary

• Base branch target: master
• Problem: Transcription was hardcoded to a single Groq endpoint — no way to use alternative STT providers
• Why it matters: Users need flexibility to choose STT providers based on accuracy, cost, or compliance requirements
• What changed: Refactored single-endpoint Groq transcription into a multi-provider architecture with TranscriptionProvider trait. Implemented five STT providers: Groq (default, existing), OpenAI Whisper, Deepgram, AssemblyAI, and Google Cloud Speech-to-Text. Added TranscriptionManager for provider routing and the transcribe_with_provider() method for explicit provider selection. Maintains full backward compatibility.
• What did not change (scope boundary): Existing transcribe_audio() function signature unchanged. Existing config fields (api_url, model, api_key) and credential resolution (GROQ_API_KEY env fallback) preserved. Callers in telegram.rs, discord.rs, whatsapp_web.rs require no changes.

Files changed

• src/channels/transcription.rs: Add TranscriptionProvider trait, five provider implementations, TranscriptionManager, shared validate_audio() helper, and parse_whisper_response() utility
• src/config/schema.rs: Extend TranscriptionConfig with default_provider and optional sub-configs (OpenAiSttConfig, DeepgramSttConfig, AssemblyAiSttConfig, GoogleSttConfig); fix pre-existing sync_directory async/sync mismatch on non-unix platforms
• src/config/mod.rs: Export new STT config types

Label Snapshot (required)

• Risk label: risk: medium
• Size label: size: L
• Scope labels: channel, config
• Module labels: channel: transcription
• Contributor tier label: (auto-managed)
• If any auto-label is incorrect, note requested correction: N/A

Change Metadata

• Change type: feature
• Primary scope: channel

Linked Issue

• Closes feat(stt): multi-provider STT with TranscriptionProvider trait #2989

Supersede Attribution (required when Supersedes # is used)

N/A

Validation Evidence (required)

Commands and result summary:

cargo fmt --all -- --check   # clean
cargo check   # passes (only pre-existing clippy warnings in unrelated files remain)
cargo test   # all 20 transcription unit tests pass (existing + new); config default, roundtrip, and without-transcription tests pass

• Evidence provided: unit test results, config roundtrip tests
• If any command is intentionally skipped, explain why: CI pipeline validation pending

Security Impact (required)

• New permissions/capabilities? No
• New external network calls? Yes — four new STT provider endpoints (OpenAI, Deepgram, AssemblyAI, Google)
• Secrets/tokens handling changed? Yes — new API key fields for each provider sub-config
• File system access scope changed? No
• If any Yes, describe risk and mitigation: Each provider's API key is optional and config-gated. Provider sub-configs default to None. Audio validation occurs before any network call. Existing Groq credential resolution unchanged.

Privacy and Data Hygiene (required)

• Data-hygiene status: pass
• Redaction/anonymization notes: Audio data sent to external STT APIs for transcription — same privacy model as existing Groq path
• Neutral wording confirmation: Confirmed

Compatibility / Migration

• Backward compatible? Yes
• Config/env changes? Yes — new optional default_provider field and provider sub-configs in [transcription] section (all default to None/Groq)
• Migration needed? No — existing configs without new fields parse correctly and default to Groq

i18n Follow-Through (required when docs or user-facing wording changes)

• i18n follow-through triggered? No — code changes only

Human Verification (required)

• Verified scenarios: Provider trait implementation for all five providers, manager routing, backward-compatible function preservation, config roundtrip
• Edge cases checked: Audio validation ordering (size/format errors before missing-key errors), missing config defaults to Groq, invalid audio formats
• What was not verified: Live API calls to non-Groq providers (requires credentials)

Side Effects / Blast Radius (required)

• Affected subsystems/workflows: Transcription subsystem, config schema
• Potential unintended effects: None — existing callers use unchanged transcribe_audio() function
• Guardrails/monitoring for early detection: Audio validation runs before network calls; provider selection explicit

Agent Collaboration Notes (recommended)

• Agent tools used: Claude Code
• Workflow/plan summary: Extracted trait from existing Groq implementation, replicated pattern for four additional providers
• Verification focus: Backward compatibility, config serde stability, test coverage
• Confirmation: naming + architecture boundaries followed

Rollback Plan (required)

• Fast rollback command/path: git revert <commit>
• Feature flags or config toggles: default_provider defaults to Groq; reverting preserves existing behavior
• Observable failure symptoms: Non-Groq STT providers unavailable (Groq continues working)

Risks and Mitigations

• Risk: New provider implementations untested against live APIs
  • Mitigation: Unit tests validate request construction and response parsing; live testing deferred to integration phase
• Risk: Config schema expansion could break existing config files
  • Mitigation: All new fields have serde(default) — existing configs parse without changes

Summary by CodeRabbit

Release Notes

• New Features
  • Transcription now supports multiple providers: OpenAI Whisper, Deepgram, AssemblyAI, Google STT, and Groq
  • Configure and select from multiple transcription providers based on your needs
  • Improved audio validation with format normalization support
  • Existing transcription configurations remain fully backward compatible