Fix assay routing and provenance CLI parity

[iskandr]

Summary

• route purified-MHC half-life measurements to binding.parquet from the structured IEDB assay method and response fields
• bump the observations artifact contract so existing cached indexes rebuild with the corrected routing
• expose one shared MHC allele provenance vocabulary and use it for the MS, binding, and training CLI choices
• bump Hitlist to 1.58.0

Source audit

The registered 2026-03-30 IEDB snapshot differs from #418's original count. PMID 36423003 has 146 cellular-MHC mass-spectrometry observations and 76 purified-MHC half-life measurements. The old classifier routed 66 of the latter to binding based on their qualitative values but leaked 10 plain-Positive rows into observations. The same structured signature identifies 942 leaked rows corpus-wide.

The IEDB source already records the modality accurately, so this is an internal Hitlist classification defect rather than an upstream data problem. The corrected counts are documented on #418.

Verification

• isolated full rebuild: all 942 matching assay IRIs moved from MS to binding; zero remain in MS
• PMID 36423003 after rebuild: 146 MS observations and 76 binding observations
• ./format.sh
• ./lint.sh
• ./test.sh — 1,198 passed
• pytest -q tests/test_build_smoke.py — 2 passed

Closes #418.
Closes #419.

[coveralls]

Coverage Report for CI Build 33894117529

Coverage increased (+0.01%) to 83.016%

Details

• Coverage increased (+0.01%) from the base build.
• Patch coverage: No coverable lines changed in this PR.
• 46 coverage regressions across 1 file.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

46 previously-covered lines in 1 file lost coverage.

File	Lines Losing Coverage	Coverage
curation.py	46	94.52%

---

Coverage Stats

Relevant Lines:	8137
Covered Lines:	6755
Line Coverage:	83.02%
Coverage Strength:	0.83 hits per line

---

💛 - Coveralls

[iskandr]

Deployed Hitlist 1.58.0 from clean main at merge commit 8fd02b3.

Release verification:

• ./lint.sh passed
• ./test.sh --all passed: 1,222 tests, one expected warning
• PyPI exposes both hitlist-1.58.0-py3-none-any.whl and hitlist-1.58.0.tar.gz
• https://pypi.org/project/hitlist/1.58.0/