After #11, we use traittypes.Array to annotate JupyterScatter.points and JupyterScatter.selection.

However, traittypes.Array is valid for any numpy array and the shape/dtype for points and selection are more strict. We should add custom validators for these attributes, which will raise an exception if an array of the wrong shape/dtype is assigned on the python side.

This isn't urgent, since jscatter.Scatter for the most part ensures the data are correct, but this would probably help with tracking down weird bugs of where dtypes aren't what are expected.