Feature: Full Support for Speaker-Diarized Data

We expect to have speaker-diarized annotations for all speech content. The system should incorporate these speaker segments throughout the entire workflow so users can search, filter, and view results with speaker context.

**Proposal:**
Add end-to-end support for diarization metadata across the system:

**1. Indexing / Storage**

- Store speaker segments (speaker ID, start/end time) in the DB.
- Link diarized segments to transcript segments and tokens.
- Ensure compatibility with multi-speaker files.

**2. Search**
- Enable filtering by speaker ID, # of speakers, or specific speaker clusters.
- Allow queries like: keyword spoken only by speaker X, keyword spoken in overlap, etc.
- Show speaker info next to each match.

**3. UI / Result View**

- Display speaker label in the snippet panel.
- Optional color-coding per speaker.
- Offer a toggle to show/hide speaker segments on the timeline.
- Add speaker breakdown stats (e.g., matches per speaker).

**4. CSV / Export**

- Include speaker ID and speaker start/end time for every result.

**Why this matters:**
Diarization makes the system more accurate (especially for multi-speaker content) and opens new research workflows: conversational analysis, speaker-specific patterns, code-switching, institutional audio, etc.

No files, tests, or entry points are named. Start by mapping the indexing/storage, search, UI/result, and CSV/export workflows, then define how speaker segments connect to transcripts and tokens; done means speaker-aware storage, filtering, display, and export across the full workflow.

Written by the indexing model from the issue text.