BB-35: New operation: GrB_subassign

[mcmillan03]

See SuiteSparse GxB_subassign function [someone fill in description here.]

[DrTimothyAldenDavis]

GrB_assign computes C(I,J)+=A, where C and M have the same size.
GxB_subassign computes C(I,J)+=A where C(I,J), M, and A all have the same size.
They are the same if I and J are both GrB_ALL, or if the mask is NULL and the null mask is not complemented.

GrB_assign can be broken down into the following steps:
(1) S = C(I,J)
(2) S = S accum A
(3) Z = C
(4) Z(I,J) = S
(5) C=Z

GxB_subassign breaks down in the these steps:
(1) S = C(I,J)
(2) S = S accum A
(3) C(I,J) = S

So far, nearly all uses I have seen of GrB_assign and GxB_subassign fall into one of 3 categories:
(1) simple case: I and J are both GrB_ALL and there is no mask
(2) just the mask is present (in many forms, complemented or not, etc), but I and J are both NULL
(3) just I and J are present but the mask is NULL.

In all those cases, GrB_assign and GxB_subassign are identical. I still think a GrB_subassign is important, and should be considered as an addition to the spec, but after writing many algorithms for LAGraph, this is not as high a priority as I once thought.

I have a section in my user guide "Comparing GrB_assign and GxB_subassign", which is section 10.10 in my v5.1.7 user guide, page 223.