Feature request: re-run transcription from History when a dictation comes out wrong

[leo-li-opus]

Motivation

Transcription/translation isn't always right — e.g. a phrase comes out garbled, or the wrong language is picked (cf. #17, Mandarin → [SPEAKING CHINESE]). Today the only recovery is to record the whole thing again and re-speak it, which is annoying for anything longer than a sentence.

Since the audio is already kept, I'd like to re-run a past dictation from the History view instead of repeating myself.

Proposed behaviour

In the History list, for an entry that still has its audio, add a Re-transcribe action that re-runs the stored audio through transcription + polish and replaces (or appends an alternative to) the transcript — without me having to record again.

Nice-to-haves on the re-run:

• Let me pick the language explicitly (override auto-detect) — this directly addresses the "translated wrong / wrong-language" case.
• Optionally pick a different model than the original.
• Show the result so I can copy/paste it, same as a fresh dictation.

Feasibility note

HistoryEntry (Sources/OpenQuackKit/History/HistoryStore.swift) already persists audioURL, language, and modelID, so the source audio and original settings are available to re-run against. Main constraint: this only works when audio history is enabled — for transcript-only entries (audioURL == nil) the action should be disabled/hidden with a hint about why.

Out of scope

• Editing the transcript text by hand (separate ask).
• Changing the live transcription pipeline accuracy itself.

[larryxiao]

Thanks, this is a clean recovery flow. The audio is already retained locally for the History view (SPEC-014), so re-running a stored clip through transcription + polish — with an explicit language and/or model override — is very doable without you re-speaking. I've added it to the roadmap as an adoption-focus idea; it also directly addresses the wrong-language case from #17. Next step is a short spec.