Add Speaker

Create a placeholder speaker on an audio file. The new speaker has zero segments — sentence ownership gets assigned via the existing transcript editor. Auto-allocates speaker_id from a monotonic counter persisted on the doc (never reuses ids freed by delete). Cached CSV/PDF reports are invalidated since the report enumerates speakers.

Returns the auto-allocated {speaker_id, name} so the FE can append the new tab without a re-read.