feat(stt): add local speech dictation with Parakeet TDT

[EduCosta85]

Description

Adds offline, privacy-first local speech-to-text (dictation) support powered by NVIDIA NeMo's FastConformer Parakeet TDT 0.6B v3 model using the sherpa-onnx inference engine.

Key highlights:

• Offline & Private: Audio is processed entirely on the user's local CPU via ONNX Runtime without requiring third-party cloud API keys.
• Multilingual Support: Supports 25 languages with automatic punctuation and capitalization.
• Dynamic Model Manager: Manages model verification (SHA-256), download progress, and caching in userData/models/stt/.
• Renderer & Settings Integration: Added "Local (Offline - Parakeet TDT)" provider under System > Voice Input settings with an integrated model downloader, status indicator, and direct voice input support in conversations.
• i18n: Added translated keys across all 13 supported languages (en-US, pt-BR, zh-CN, zh-TW, ja-JP, ko-KR, es-ES, fr-FR, de-DE, ru-RU, uk-UA, tr-TR, fa-IR).

Related Issues

Type of Change

• fix — Bug fix (non-breaking change which fixes an issue)
• feat — New feature (non-breaking change which adds functionality)
• perf — Performance improvement
• refactor — Code restructuring (no behavior change)
• Breaking change (fix or feature that would break existing functionality)
• docs — Documentation update

Atomic PR Checklist (Rule 1)

• This PR contains exactly one feature or bug fix that cannot be further decomposed
• The PR title follows Conventional Commit format: <type>(<scope>): <subject> (English)

Local Checks (Rule 3)

• bun run format — formatting passes
• bun run lint — no lint errors (skip if no .ts/.tsx changed)
• bunx tsc --noEmit — no type errors (skip if no .ts/.tsx changed)
• bunx vitest run — tests pass
• i18n validated (bun run i18n:types + node scripts/check-i18n.js) — only if src/renderer/, locales/, or src/common/config/i18n/ changed; N/A otherwise
• New/changed user-facing text uses i18n keys (no hardcoded strings)

Runtime Verification

• Verified on macOS
• Verified on Windows
• Verified on Linux
• I have performed a self-review of my own code

Additional Context

Tested and verified with real spoken audio on macOS (Apple Silicon arm64). Inference time is ~1.4s on CPU with accurate punctuation and capitalization.

[codecov]

Codecov Report

❌ Patch coverage is 73.49398% with 88 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...esktop/src/process/services/speech/modelManager.ts	71.42%	20 Missing and 12 partials ⚠️
...nts/SystemModalContent/VoiceInputSection/index.tsx	65.07%	11 Missing and 11 partials ⚠️
.../src/process/services/speech/localSpeechService.ts	80.55%	7 Missing and 7 partials ⚠️
...sktop/src/renderer/services/SpeechToTextService.ts	45.83%	9 Missing and 4 partials ⚠️
...esktop/src/process/services/speech/sherpaLoader.ts	84.61%	4 Missing ⚠️
packages/desktop/src/process/bridge/index.ts	0.00%	1 Missing ⚠️
...ackages/desktop/src/process/bridge/speechBridge.ts	91.66%	0 Missing and 1 partial ⚠️
...src/renderer/services/speech/speechStreamPolicy.ts	75.00%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!