Science spot checks on the 64-tile Nibi run before scaling to 1000 deg² #884

The redesigned orchestration (#852) plus the new mask handling (#845/#846) and PSF-star selection policy changed a lot at once. Before producing 1000 deg², check the 64 tiles already run on Nibi at the science level: PSF-star selection and outlier-rejection statistics (train/test splits and rejection stats are saved), kept/rejected star plots per CCD and tile, mask before/after comparisons, and basic catalogue sanity. Anything found to be missing from the retained outputs goes to #844.