TST: dropping of nuisance columns for groupby ops

[rhshadrach]

xref #38774

As far as I can tell, for groupby we only have two tests for dropping nuisance columns and these only involve sum/mean. A more robust and parameterized test over all the groupby ops that drop nuisance columns should be added.

[horaceklai]

take

[horaceklai]

Is there a list of aggregation functions where nuisance columns are dropped?

I expected functions such as max, min, last, first to drop columns of object datatypes. But I was wrong.

So far I have mean, sum, std, var, sem