preprocessing

Shape-based preprocessing for the cst_calculation v1 step.

This step consumes two RAW cached frames: retinal-layers segmentation output and fovea v7 landmark output. A single preprocess serves both — it interrogates each frame's shape and applies the matching pipeline (retinal-layers->segmentation or fovea v7->v6), rather than matching on the producing model's ref/version. Unknown shapes pass through unchanged.

---