[manhnguyen24-dev]

What changed

• Add push-to-talk recording beside Send with editable transcript insertion and cancellation.
• Add configurable STT provider profiles, model catalog loading, ordered fallback, and local/private routing.
• Add an optional on-demand multilingual Local Whisper runtime with platform asset manifests and a build workflow.
• Document the feature and API, with focused frontend and backend coverage.

Verification

• Focused voice UI tests: 6 passed.
• Backend catalog/runtime tests: 5 passed.
• ESLint and Ruff checks passed; Git whitespace check passed.
• Live Groq STT smoke test returned HTTP 200 and a transcript from a synthetic sample.