feat: Meeting Domain prompt support for Whisper (vocabulary hints via initial_prompt)

[Polsemaker]

Description

Adds a "Meeting Domain" feature that lets users provide vocabulary hints (custom names, jargon, technical terms) to bias Whisper transcription via its initial_prompt parameter.

Domains are plain-text files in ~/Library/Application Support/com.meetily.ai/domains/ (falls back to ~/.meetily/domains/). When a domain is active, its contents are forwarded as initial_prompt to whisper-rs at every transcription call site. No domain selected → unchanged behavior.

UX surfaces:

• New "Meeting Domain" section in Transcript Settings with a dropdown + a Manage… dialog (list / edit / delete / open folder / char counter)
• A "Domain" button in the transcript header next to the existing Language button, so users can switch mid-recording
• A "Meeting Domain" dropdown in the Retranscribe dialog with a "None" option, so users can A/B different hint files without changing their global setting
• All Domain UI is hidden when Parakeet is the active provider (Parakeet doesn't support initial_prompt)

Implementation notes:

• New module: frontend/src-tauri/src/meeting_domain/mod.rs (file CRUD, name validation, ~1000-char / ~224-token truncation matching whisper.cpp's hard limit)
• Global MEETING_DOMAIN state mirrors the existing LANGUAGE_PREFERENCE pattern; Tauri commands surface get/set/list/save/delete to the frontend
• initial_prompt is threaded through transcribe_audio / transcribe_audio_with_confidence so every whisper invocation receives the same hint (live worker, standalone command, parallel processor, audio import, retranscription)
• ConfigContext persists selectedDomain to localStorage and syncs to Rust on mount, mirroring selectedLanguage

Related Issue

Fixes #474

Type of Change

• New feature
• Bug fix
• Documentation update
• Performance improvement
• Code refactoring
• Other

Testing

• Unit tests added (7 tests in meeting_domain::tests covering name validation, listing, save/delete round-trip, and truncation)
• Manual testing performed on macOS with NB-Whisper-large (Norwegian) — confirmed proper nouns and domain jargon are recognized correctly when the matching domain file is active, and that unchecking the domain restores baseline behavior
• cargo test meeting_domain — 7/7 pass
• cargo build — clean (warnings are pre-existing in unrelated files)
• tsc --noEmit — 0 errors

Documentation

• Documentation updated
• No documentation needed for initial review — happy to add a docs page once the design is accepted

Checklist

• Code follows project style
• Self-reviewed the code
• Added comments for complex code
• Updated README if needed
• Branch is up to date with devtest
• No merge conflicts

Screenshots

Settings → Transcription → Meeting Domain (with Manage dialog)

Transcript header — Domain button next to Language

Retranscribe dialog — per-retranscription Meeting Domain override

Additional Notes

• ~224 token (~1000 char) cap on initial_prompt is enforced with a graceful truncation + log warning, matching whisper.cpp's hard limit.
• Domain UI is provider-aware: hidden for Parakeet to avoid suggesting a no-op control.
• Selected domain is read fresh at each transcription start, so editing a domain file mid-session takes effect on the next recording without a restart.
• Design intentionally orthogonal to [FEATURE] VibeVoice-ASR for speaker diarization and domain-specific terminology #335 (VibeVoice-ASR) — this is a small biasing knob on the existing Whisper path, not a new engine.