[FIX] stats: Handle empty array

[VesnaT]

Issue

Fixes #6208 (comment)

Description of changes

Compute stats for an empty array / Table.

Includes

• Code changes
• Tests
• Documentation

[janezd]

@VesnaT, I don't think this is correct fix. I'd even dare putting assert X.ndim == 2 here. I think we should always safely assume that X is two-dimensional, where X.shape[1] equals len(data.domain.attributes).

The problem is probably somewhere in class Table, which - as far as I see for now - constructs a zero-dim array when selecting zero rows. The problem must be fixed there; unless we prevent zero-dim arrays, everything will crash.

[markotoplak]

@janezd, I dare to say you forgot about our beautiful data.Y, that can certainly be 1-dimensional (and plays the role of X here).

[janezd]

@markotoplak, you're wrong. I didn't forget that; at the time of writing I haven't gone deep enough yet, and I thought this is just about X (as in data.X). I didn't yet know this X could also be our beautiful shape-shifting Y. :)