Add structured speaker editing, view, and DOCX export - #1587

Summary

• persist diarization results in a dedicated speaker field instead of embedding labels into transcript text
• add an editable Speaker column to the timestamp editor, with inline choices, custom names, and multi-row assignment
• add a Speakers view with stable colors, explicit names, consecutive-turn grouping, and per-speaker filtering
• add DOCX – Speakers export with matching colored/bold speaker labels, normal black transcript text, and optional timestamps
• retain speaker context in TXT/SRT/VTT exports without polluting the stored transcript text
• safely migrate existing databases without heuristically parsing legacy Name: text prefixes

Behavior

Identify Speakers now saves clean transcript text and structured speaker metadata. The Speakers view and DOCX export share the same grouping and color model, so the on-screen result and Word document remain consistent. Both features are available only when at least one speaker is assigned.

In the timestamp table, users can edit a speaker per row, create a new speaker name, or assign one speaker to multiple selected rows. The generated Word document is a normal, editable .docx file.

Motivation

This completes the editing and export side of the known-speaker workflow discussed in #1043 and complements #1586. Importantly, the structured speaker model is independent of the diarization backend: MSDD, Sortformer, or a future model can all populate the same field and use the same editor, view, filters, and exports.

That makes this useful beyond the current MSDD speaker-count implementation, especially for interviews, meetings, hearings, and court-style transcripts that need review and document export after diarization.

Additional UI fixes

• keep the audio control area compact so the transcript table uses the available height
• reset rich speaker-label formatting when returning to the plain Text view

Testing

• 53 passed, 30 skipped in the focused schema, editor, identification, export, and DOCX plugin suites
• database migration test confirms existing transcript text is preserved
• generated DOCX XML is validated and checked for speaker names, colors, and optional timestamps
• user validation on a 30-minute Hungarian room recording with three participants
• manual Windows testing confirmed speaker editing, filtering, and the colored Word export