BUG: fix incorrect output unit for np.prod with an axis argument

[kyleaoman]

Closes #536

When an axis keyword is passed to np.prod, the power that the dimensions should be raised to are not the size of the array but the ratio of the sizes of the input array and returned array (it's always this, but without the kwarg the size of the returned array is 1).

[kyleaoman]

pre-commit.ci autofix

[neutrinoceros]

Thanks for noticing and patching this !
The same problem is present with np.nanprod, and your patch also fixes it, however that's mostly due to an implementation detail in numpy that we don't control so it should be tested too.
Otherwise, LGTM

[kyleaoman]

@neutrinoceros done - just parametrized the new test across prod and nanprod.

[neutrinoceros]

Thanks ! let's get this one in !