prod and prod_axis tensor operators

[wbrickner]

Feature description

Product of all elements in a tensor, and those along a given axis, just like we have sum and sum_axis.

Feature motivation

Basic tensor operation missing. I personally need it for some ML research.

(Optional) Suggest a Solution

Current workaround is tensor.log().sum().exp(), but this is slower and should not be required of the end user.

---

It would also be nice, although it is a separate matter, to have function to sum/prod along multiple axes at once in parallel. Otherwise you need to do two sequential tensor operations which reduces parallelism utilization.

For instance, a 3D tensor could be summed along its first two axes, perhaps like tensor.sum_axes([0, 1]);, or tensor.prod_axes([1, 3]).

[antimora]

Linking someone's attempt to implement this: #1173