As a user, I can provide a speaker count range so that diarization has better constraints

[julien731]

User Story

As a user, I can provide a min/max speaker count range during upload so that diarization has better constraints without requiring me to know the exact speaker count.

Spec: docs/specs/transcription-quality-improvements.md — US-4

Size: S

Acceptance Criteria

• The upload form includes optional "Min speakers" and "Max speakers" fields
• Given an exact speaker count is provided, when min/max are also set, then min/max are ignored
• Given only min speakers is provided, when diarization runs, then at least that many speakers are preserved
• Given no speaker hints are provided, then default diarization behavior is unchanged
• min_speakers and max_speakers fields are persisted in MeetingMetadata
• Given min_speakers exceeds the actual number of speakers, then diarization may split one speaker into multiple (acceptable tradeoff)

Truths

• T1: min_speakers/max_speakers are mutually exclusive with num_speakers — exact count takes precedence (BR-2)
• T2: These values are pass-through to pyannote's diarization pipeline

Business Rules

• BR-2: min_speakers/max_speakers are mutually exclusive with num_speakers. If exact count is set, range is ignored. Pyannote does not accept both simultaneously.

Edge Cases

• min_speakers > actual speakers → diarization may over-split (acceptable tradeoff vs. merging)
• Both exact count and min/max provided → exact count wins, min/max ignored

Dependencies

None