Expanded support for NEP 13 and 18

[henryiii]

I think it might be useful to have expanded support for NEP 13/18. I was just looking at making a cumulative histogram. Currently, you can do that with:

array: NDArray = np.cumsum(h.values())
# Now you have to make the histogram

And we can even make this shorter:

array: NDArray = np.cumsum(h)
# Still have to make the Histogram, even though you are starting with a Histogram

But this still returns an array. Ideally, I think it should keep the type, that is, return a Histogram. Using NEP 18, this works:

    def __array_function__(self, func, types, args, kwargs):
        if func == np.cumsum:
            h = self.copy()
            h[...] = func(h.view(), *args[1:], **kwargs)
            return h
        return NotImplemented

...

cumsum_hist: bh.Histogram = np.cumsum(h)

Would this make sense to implement, possibly with arbitrary functions allowed? It would defer to the View, which then would do whatever it supports (also via NEP 13/18), and then produce a new Histogram with axes, etc. carried over, but with the new contents.