PR summary

Closes #8108 (which was partially addressed already in #8224)

The 10 candidate legends are the same size and grid aligned, and so share edges. We can speed up the count_contains hot path by computing whether each point lies in those candidate intervals once, rather than looping over each location. In my profiling, I see a ~5x speedup for Legend._find_best_position, and most of the time there now is spent in other operations.

Did not deprecate count_contains since I found some instances of it being used on github, but it's no longer used anywhere internally so could remove. Added a test since it's no longer being exercised otherwise.

Before:

After:

Should hopefully cut down a lot on the warning Creating legend with loc="best" can be slow with large amounts of data, which I seem to run into fairly often.

AI Disclosure

Claude code used for first cut, manually cleaned up and refactored.

PR checklist

• "closes #0000" is in the body of the PR description to link the related issue
• new and changed code is tested
• [n/a] Plotting related features are demonstrated in an example
• [n/a] New Features and API Changes are noted with a directive and release note
• Documentation complies with general and docstring guidelines