pipelines

Centralised preprocessing pipelines (ordered lists of typed transforms).

Raw model inference is cached unprocessed (see model_inference step), so the format-specific transforms a consuming step needs are applied just-in-time inside that step. A pipeline is an ordered list of typed PreprocessingConfig; the transforms it should run against a given frame are selected by a shape detector (see bitfount.preprocessing.detectors).

The pipelines live here — rather than inside a single step — because more than one step shares them (the pathology-to-segmentation pipeline is used by both GA calc steps and the PDF report).

---