Fix default size for points elements

[LucaMarconato]

When points elements are being shown, an optimal size for the points is guessed based on a heuristic implemented in _calc_default_radii().

There are three fixes that we need to make here:

• get_extent(), even when using exact=False, is too slow when there is a large number of points. We should add a parameter fast: bool = False which computes a very rough approximation of the extent based on some "Montecarlo" sampling.
• The value in unit we return is wrong when we are in a scaled coordinate system; I think that we need to divide by the average module of the eigenvalues of the affine transformation, as we do in _adjust_radii_of_points_layer()
• There is a strange bug in get_extent() where the single-dispatch doesn't recognize SpatialData objects. This is tracked here: get_extent() does not recognizes SpatialData in the single dispatch pattern in specific cases spatialdata#500

For the moment, I am disabling the use of _calc_default_radii() in #215.