This uses #481 (and is actually why I made #481).

I still need to write the docstring.

I tried to make this efficient by performing repeated squaring. For example, for A.power(8), this will compute:

cur = op(A @ A).new()
cur << op(cur @ cur)
updater << op(cur @ cur)

This makes the implementation a little... uh... not obvious, but it works well.

A.power(n) returns an expression so mask can be applied. Even if a mask is used, an intermediate result A.power(n // 2) can still be pretty large--we only apply the mask on the very last matrix multiply.

I decided to require n to be a positive integer. n == 0 could return "the" identity, but what would be the identity for different semirings?

This method will let us easily and efficiently implement number_of_walks in graphblas_algorithms.