BUG: Power override for zero power

[JBorrow]

Implements a fix to #203.

Comments welcome as this could be seen as quite a dirty fix, as we're taking away ufunc dispatch from numpy.

[ngoldbaum]

With this PR if you take a Unit object to a zeroth power, do you get back dimensionless? I suspect you don’t since the content of __pow__ is unchanged. Since numpy had an optimization that shortcuts to ones_like you may need to make changes both to the unyt_array class and to the Unit class to make their behavior consistent. I would also like to see a test for the behavior of the Unit class independent of the behavior of the unyt_array class. I would also move the comment change to the unit object’s __pow__ method to the new method you added to unyt_array. The unit object doesn’t know anything about numpy’s ufunc system.

[JBorrow]

At the moment the Unit object happily goes to the zeroth power (I've added a test for this, though). The unit calculation internally is completely fine (and always has been); sympy handles the 0th power gracefully.

The base issue here is that we have the _ones_like ufunc mapped to _preserve_units (which makes sense,

unyt/unyt/array.py

Line 512 in e86c050

_ones_like: _preserve_units,

). Unfortunately this same ufunc is used for behaviours that aren't consistent when units are associated with the data.

I've moved around the comments as requested.

[ngoldbaum]

Thanks for clearing up where the issue was and updating the tests and comments!