UnitArray __rmul__ and __rdiv__ drop units when isinstance(other, numpy.generic) about scimath

[mdickinson]

So the issue is really in numpy: if c is of type numpy.float64, then c.__mul__ incorrectly accepts a UnitArray instance and returns a result without units, rather than delegating responsibility for the multiplication to UnitArray.__rmul__ as we'd like it to.

The only way I see to fix this in scimath.units would be to not have UnitArray inherit from ndarray any more, but that's a bit of a drastic move that would almost certainly break a lot of existing code. [From a purity point of view, I think that it's actually the right move: a UnitArray is a product of an array with a unit, and for me it's a stretch to regard that as an ndarray in its own right. I remember seeing similar difficulties from people trying to create a 'Money' class that inherited from the standard library 'Decimal' type.]

[rupertnash]

I think this isn't the whole story though. I tried running c*a under pdb and the __array_wrap__ method is called afterwards. My understanding of the numpy API is that this method is called on the object involved which has the highest __array_priority__ attribute. So this method might be suitable for sorting out the units?