Sub-issue: speaker diarization

## What
Add speaker diarization so transcripts can separate who said what.

## Why
Without diarization, meeting transcripts are harder to read and summaries lose context. This is one of the main feature gaps compared with Meetily.

## How
- define a diarization step in the Rust backend
- detect speaker segments from recorded audio
- attach speaker labels to transcript chunks
- render speaker-aware transcript UI in the frontend
- make it optional so local-only installs still work well on modest hardware

Parent: #51

Start by reading parent issue #51 and tracing the existing Rust backend transcription flow and frontend transcript rendering. Define how recorded-audio speaker segments map to transcript chunks, then verify optional behavior for local-only installs and speaker-aware rendering across both areas.

Written by the indexing model from the issue text.