Numpy2 compatibility (brian2.units)

[mstimberg]

Various fixes for to make the unit system compatible with the upcoming numpy 2.0 release. It might make sense to revisit this more fundamentally at a later point, but for now the main change is that __array_wrap__ and __array_prepare__ are merged into __array_ufunc__. Without doing anything further, this actually enables some new functionality like:
Before:

>>> np.add.reduce(np.arange(3)*mV)
Quantity(0.003)

After:

>>> np.add.reduce(np.arange(3)*mV)
3*mV

But I don't think many users do complex calculations with Quantity arrays in the first place. Still, we need to point this out clearly in the release notes, it might break some code. As discussed, the changes in numpy's system also lead to things like x**(2*volt) no longer raising an error, since float(2*volt) == 2.0 and this makes things go through the square instead of the pow ufunc.