[Sad Catapella B4][Performance] Mask-looking filename ownership checks scale quadratically with image count #84

Baseline and classification

Sad Catapella Batch 4, commit d7a538e56d41ac1093ba79adcad397c21b3dfd32. Confirmed asymptotic repeated work, speed magnitude not measured.

Evidence

• discover_experiment() lines 409–415 invokes mask_name_issues for every discovered dataset by default, even without a masks folder.
• mask_name_issues() lines 453–467 runs the image-name check and mask-folder ownership check across all images.
• images_named_as_masks() lines 529–539 invokes _mask_owner() lines 564–583 for each eligible image; _mask_owner reconstructs keys by scanning every stem in the experiment, so ordinary N-image datasets incur O(N²) prefix comparisons. Mask-like folder checks add further calls.
• The ownership rule is longest matching position name, prefer same folder, which must remain deterministic and unchanged.

Optimization acceptance

Use direct candidate lookups from known suffixes for default mask-name detection; for wildcard ownership in a masks-like folder, use a prefix trie or a carefully indexed prefix search to find the longest valid name, retaining same-folder priority. Reuse the case-normalized stem/position index once; avoid repeated position_id and PurePosixPath construction. Benchmark 1k/10k/100k positions, long shared prefixes, multiple extensions, cross-condition identical stems and worst-case mask-like names. Assert identical warnings and owner references against the existing algorithm on an independent fixture. No Numba: Python indexing is the right tool.