Raising to negative power does not flip errors as expected

[cgobat]

In #8, @sevenstarknight brought to light the fact that the behavior of power operations with exponents < 0 is not consistent with the equivalent division operation. In other words, given x = a_u(...), 1/x currently produces a different result than x**(-1.0). These should, of course, be equivalent operations.

The implementations of the __pow__ and __rpow__ methods should be investigated and adjusted accordingly.