fix(estimator): pass distance metric parameters as metric_params to NearestNeighbors

[TasfinMahmud]

Fixes #1390.

NearestNeighbors expects metric parameters (like the V matrix for Mahalanobis distance) to be passed through the metric_params argument, rather than as top-level kwargs.

This PR passes self.distance_metric_params using the metric_params argument in DistanceMatchingEstimator. A regression test is included to ensure method_params={"V": ...} behaves correctly end-to-end without failing.

[github-actions]

🤖 This is an automated response from Repo Assist.

Thanks for contributing this fix, @TasfinMahmud! The metric_params= change is exactly right — passing Mahalanobis parameters via **kwargs at the top level of NearestNeighbors has never worked, and this one-liner correction unblocks the feature.

Context for maintainers: Repo Assist previously submitted PRs #1475, #1508, and #1526 for this same issue (#1390). All three have now accumulated merge conflicts since PR #1465 (which refactored exact_match_cols logic in the same file) was merged. This contributor PR is rebased on current main and cleanly supersedes the stale Repo Assist PRs — I'd recommend closing #1475, #1508, and #1526 in favour of this one.

One minor note: PR #1526's fix included metric_params=self.distance_metric_params if self.distance_metric_params else None (passing None when the dict is empty) while this PR passes the empty dict directly (metric_params={}). Both are equivalent for sklearn, so this is not a blocker.

The fix and test look good. Once CI runs and passes, this should be merge-ready.

Generated by 🌈 Repo Assist, see workflow run. Learn more.

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@11c9a2c442e519ff2b427bf58679f5a525353f76