map/map! for sparse matrices

[Sacha0]

This pull request provides (1) a map/map! implementation specialized for a single (input) sparse matrix, (2) a map/map! implementation specialized for a pair of (input) sparse matrices, and (3) a general map/map! implementation capable of efficiently handling an arbitrary number of sparse matrices (within compiler limits). broadcast/broadcast! over a single (input) sparse matrix being identical to map/map! over the same, this pull request also reimplements broadcast/broadcast! over a single (input) sparse matrix as short children of map/map!.

The semantics of these map!/broadcast! methods are those discussed in #19372. Notably these map!/broadcast! methods only allocate when strictly necessary, so for example

julia> C = spzeros(5,5); A = speye(5); B = sparse(Bidiagonal(zeros(5), ones(4), true));

julia> @allocated map!(*, C, A, B) # after warmup
0

Additionally, where these map!/broadcast! methods do allocate, they do so less pessimistically than the existing broadcast! methods.

The specialized map/map! methods achieve performance similar to corresponding existing broadcast/broadcast! methods (despite performing slightly more work to enable the more precise allocation behavior). The former win in some cases and the latter in others, with runtime differences usually ~10% or less in (hopefully realistic) tests.

The general map/map! methods achieve performance not too far behind the methods specialized for pairs of (input) sparse matrices. The latter typically best the former by only 10-35% runtime, though in some cases they're on par and in others the gap stretches to 45%.

Fixes (the SparseMatrixCSC part of) #19363. Also relevant to #16285 and friends as a step in the sparse broadcast/broadcast! overhaul. Best!

(Realized this PR lacks a map!(f, A) method. Punting to a followup PR.)

[Sacha0]

The abstractarray test failure should be fixed.

[Sacha0]

Travis macOS hit an openblas-related error seemingly similar to that in #19379?

[Sacha0]

Absent objections or requests for time, I plan to merge this Friday morning PST. Best!