fix(sql): use comparison coercion for IN lists

[Th0masclassic]

/claim #1659

Summary

Fix a remaining static IN (...) semantic gap from #1659: compatible values that Dozer's normal equality operator coerces were compared as raw Field variants inside IN lists.

For example, current main evaluates 42 = 42.0 as true using SQL comparison coercion, while 42 IN (42.0) evaluates to false.

This patch reuses the existing equality evaluator for non-null list comparisons so IN and NOT IN are consistent with Dozer's normal equality semantics.

This is intentionally scoped separately from the NULL-semantics work in #2494 and from the IN (SELECT ...) implementations. Existing NULL behavior is preserved.

Changes

• Reuse evaluate_eq for compatible non-null IN list items.
• Add regression coverage for 42 IN (42.0).
• Add regression coverage for 42 NOT IN (42.0).
• Preserve existing NULL behavior.

Demo

Short demo video: https://github.com/Th0masclassic/dozer/releases/download/dozer-1659-in-coercion-demo/demo.mp4

The demo shows the upstream semantic mismatch, the patched behavior, and the validation results.

Validation

• Regression test confirmed to fail against the upstream implementation.
• Regression test passes with this patch in GitHub Actions and locally.
• Existing IN expression tests pass locally: 3 passed, 0 failed.
• rustfmt --check passes for the modified Rust files.
• git diff --check passes.
• CI: https://github.com/Th0masclassic/dozer/actions/runs/34276426474

AI-assisted development disclosure: ChatGPT was used to assist with investigation and implementation; the resulting changes and regression coverage were reviewed and validated before submission.

[CLAassistant]

All committers have signed the CLA.