Fix remaining open issues: typos, resource leaks, Hearst pattern, scorer

[Chessing234]

Summary

Addresses several small open bugs reported in issues:

• fix: typo 'constuctor' → 'constructor' in CandidateGenerator error message #601 — Fix typo in CandidateGenerator docstring (constructor)
• fix: truncate cache filenames to avoid OSError on eCryptfs/encfs filesystems #590 — Truncate URL/ETag hashes in cache filenames for eCryptfs/encfs path length limits
• fix: misleading variable name in create_tfidf_ann_index — number_of_non_empty_tfidfs counts empty vectors #593 — Rename misleading number_of_empty_tfidfs variable
• fix: which_look_like Hearst pattern uses hyponym twice instead of hypernym #595 — Fix which_look_like Hearst pattern to end with hypernym
• fix: unclosed file handles in read_full_med_mentions() and read_ner_from_tsv() #599, fix: unclosed file handle in evaluate_ner() when dumping metrics #597 — Use with open() for file handles in data_util.py and train_utils.py
• fix: set-list zip produces incorrect span pairings in PerClassScorer (candidate_generation.py) #592 — Fix PerClassScorer span pairing (iterate predicted spans directly)

Test plan

• Run existing scispacy test suite
• Spot-check entity linking with cached model downloads on encfs/eCryptfs

Made with Cursor