`np.fmod` doesn't work with negative inputs

[benwilliamgraham]

The current implementation doesn't follow numpy's fmod implementation which states:

The remainder has the same sign as the dividend x1

Ex:

>>> np.fmod(-4, 10)
-4
>>> numba.jit(lambda: np.fmod(-4, 10))()
2

[gmarkall]

Thanks for the report, I can reproduce this behaviour.