Automatic inference for semantic legends for marker/scatter size

[cvanelteren]

How about this @gepcel ?

This PR makes sizelegend line up with scatter marker sizing without forcing users to manually reconstruct UltraPlot’s size-scaling rules. Scatter collections now retain their effective size-scale metadata, and sizelegend() can reuse that information by default to infer matching legend marker sizes from an existing compatible scatter plot, while still allowing explicit overrides and preserving direct area=True / area=False behavior when requested. The change also adds regression coverage for absolute- size, scaled-size, compatibility fallback, and explicit override cases so the legend behavior stays predictable.

[codecov]

Codecov Report

❌ Patch coverage is 93.33333% with 8 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
ultraplot/legend.py	87.50%	4 Missing and 4 partials ⚠️

📢 Thoughts on this report? Let us know!

[cvanelteren]

@gepcel

ax.plot(..., m='.', ms=50) creates a Line2D point with marker '.' and markersize=50, but your sizelegend([50, 25], area=False) is building semantic legend entries with the default marker 'o'. Numerically the first legend handle is 50 pt, but it is marker='o', not marker='.', so it renders differently and looks wrong for that plot.

[gepcel]

@gepcel

ax.plot(..., m='.', ms=50) creates a Line2D point with marker '.' and markersize=50, but your sizelegend([50, 25], area=False) is building semantic legend entries with the default marker 'o'. Numerically the first legend handle is 50 pt, but it is marker='o', not marker='.', so it renders differently and looks wrong for that plot.

Oh, despite using matplotlib for a long time, I didn't realize '.' and 'o' are different markers. I mistakenly thought they were the same. My bad! Sorry to bother.