Implement np.gradient

[holymonson]

Feature request

Implement np.gradient, it mainly depends on np.diff.

---

A simplest overload:

@numba.extending.overload(np.gradient)
def np_gradient(f):
    def np_gradient_impl(f):
        out = np.empty_like(f, np.float64)
        out[1:-1] = (f[2:] - f[:-2]) / 2.0
        out[0] = f[1] - f[0]
        out[-1] = f[-1] - f[-2]
        return out

    return np_gradient_impl

[esc]

@holymonson thanks for submitting this request to the Numba issue tracker. This is a duplicate of: #4074 so I will be closing this in favor of that PR. Feel free add some tests and submit the suggested overload as a PR.