Vectorize check_conditional_requirement (~127x speedup)

[sstruzik]

Summary

• Replace the per-row df.apply(check_cr, axis=1) in Validator.check_conditional_requirement with column-wise boolean masks. Per trigger column, a single vectorized (~blank) & non_default & truthy mask is computed once; per required field, the OR of the masks of triggers that demand it is intersected with that field's blank mask.
• The remaining row-wise apply runs only over the subset of rows that actually have a missing CR field, just to build the comma-joined label string.
• Behavior parity preserved: numeric 0 does not trigger, blank cells do not trigger, cells equal to the schema Default do not trigger, and a required field whose column is entirely absent is reported by its field name (matching the original).

Performance

Measured against exposure (location 9501x80, account 221x41):

Step	Before	After
check_conditional_requirement	7.239 s	0.057 s
TOTAL checks	7.866 s	0.756 s

Test plan

• pytest tests/test_ods_package.py -v (50 passed)
• Run ods_tools check against an exposure that intentionally trips conditional-requirement errors and diff the messages against the prior implementation row-by-row

🤖 Generated with Claude Code