[Stretch] Add language auto-detection for transcription

**Why:** Users may share audios in different languages; auto-detect instead of manual selection.

**What to build:** Use Transformers.js language detection model (e.g., facebook/mms-lid) to identify audio language before transcription. Pass detected language to Whisper. Show detected language in UI for user confirmation.

**Acceptance criteria:**
- [ ] Language detection model loaded (e.g., facebook/mms-lid)
- [ ] Model runs on audio before transcription
- [ ] Detected language passed to Whisper transcribeAudio capability
- [ ] UI shows "Detected: Spanish" with option to override
- [ ] Fallback to profile language if detection fails
- [ ] Test with multi-language audio samples

**Depends on:** #13

Start by reviewing the implementation and interface established by dependency #13, especially the transcribeAudio capability. Trace where audio reaches transcription, then identify the Transformers.js integration point, the UI language-selection flow, and existing test or sample-audio entry points. Done means detection, Whisper language passing, confirmation UI, fallback behavior, and multilingual samples are covered.

Written by the indexing model from the issue text.