Improved datetime handling in utils to avoid type mixing

[poplarShift]

I came across two util functions that do not take into account the possibility that different datetime types may be present in an array, such as when comparing axis ranges in overlayed plots (how this can happen in the first place might be worth investigating in more detail at some point). E.g. combining both pandas and numpy datetimes in the same array ruins calls such as to numpy.max etc.

Fixes #3528

[philippjfr]

Can you think of any tests you could add here? Otherwise I'm happy to merge this.

[poplarShift]

Thanks, I'll think about that. Don't think we'd want to be too strict on what it should be able to handle either.

[philippjfr]

Merging this despite notebook tests failing, the regression is due to changes in param which are fixed in #3474.