[Feature]: Re-transcribe a saved recording (optionally with a different model or language)

Problem

If a transcription comes out poorly — wrong language auto-detection, a small model, noisy audio —
there is no way to re-run just that recording. The user must delete and re-record/re-import.

Real examples from the Library: Dutch/German/English mixed in one list; short clips that produced
"Just"; a German clip mis-detected as English.

Desired

From a saved recording, offer Re-transcribe… with:

• Model selection (default = the model that produced the existing transcript).
• Optional forced language (override auto-detect).
• Clear progress and error handling (depends on [Robustness] Model download & switch: cancellable lifecycle, integrity checks, and failure recovery #102 robustness).

Open questions

• Replace the existing transcript, or keep versions for comparison?
• What happens to downstream AI enhancement / generated notes built from the old transcript?
• Rate/perf considerations for long media.

Where it would live

• Library row context menu: Pindrop/UI/Main/TranscriptionHistoryRow.swift:284
• Detail view context menu: Pindrop/UI/Main/MediaTranscriptionDetailView.swift:796

Audio is retained in MediaLibrary, so re-running is feasible.

Notes

The existing "offline re-transcription" is an internal streaming-finalize pass, not a user action.
Related: #68 (auto-detect language hallucination).