feat(transcribe): Expose quantisation subdivision in transcribe() and CLI #309

build_score() already supports a quantization_subdivision parameter (defaulting to 16 for 16th notes), which snaps onset times to the nearest musical grid position. However, neither transcribe() nor the CLI expose this — it's always hardcoded to 16.

Currently the only way to use a different grid is to call the pipeline steps manually and pass quantization_subdivision directly to build_score().

Requested change: Add an optional quantization_subdivision parameter to transcribe() (defaulting to 16) and thread it through to build_score(). Add a corresponding --quantize CLI flag (e.g. --quantize 8 for 8th notes). This would let users control score resolution without bypassing transcribe().