[Tier 4] Apply cross-speaker reconciliation to single-file / crash-recovery run() path

[fmasi]

Symptom: Speakers are inconsistent across segments of a single-file / crash-recovery recording.

Root cause

TranscriberApp/Services/TranscriptionRunner.run() performs no cross-segment speaker reconciliation. Multi-segment crash-recovery recordings get independent per-segment labels (related to the per-segment dedup structure around :72-105).

Proposed fix

• Reuse the same identity/reconciliation logic across recovery segments so speakers are consistent in the single-file path too.

Acceptance criteria

• Speakers consistent across recovery segments.