Questions that could potentially be answered by the dataviz evaluation project

* Defaults for 2d kdes and plot_pair -> scatter vs kde vs hexbin.
* Should it depend on the number of samples to plot?
* Or should the default always be the same but some types like scatter should be discouraged if we have >10k datapoints?
* For 2d kde in general, is the density levels default good? Or should the default be levels defined by the probability within them maybe also with a [label in the contour line](https://matplotlib.org/stable/gallery/images_contours_and_fields/contour_label_demo.html)?

Start with the dataviz evaluation project and the linked Matplotlib contour-label example. Compare candidate defaults for 2D KDE and plot_pair across scatter, KDE, and hexbin, including sample-count behavior and density-level labeling. Done means the evaluation records a decision for each question.

Written by the indexing model from the issue text.