Review docstring and possibly raise a userwarning for pearsonr based on preserve_dims=all

[tennlee]

Issue: pearsonr may be mathematically undefined when all dimensions are preserved, resulting in NaNs being returned. This isn't exactly wrong, but is perhaps not what a user would be expecting and could be made clearer

Solution: This should at least be documented, and possibly a user warning should also be raised in the code

See also: KGE uses pearsonr in its calculation so is also affected - it has this described in the docstring