Hoist eltype calls in matrix multiplication

[charleskawczynski]

This PR hoists eltype calls in the matrix multiplication.

main

main
julia> using Revise; include(joinpath("test", "MatrixFields", "matrix_field_broadcasting.jl"))
[ Info: getidx times max(interior,left,right) = (3 nanoseconds,3 nanoseconds,3 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,4 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,5 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,4 nanoseconds)
[ Info: getidx times max(interior,left,right) = (7 nanoseconds,7 nanoseconds,7 nanoseconds)
[ Info: getidx times max(interior,left,right) = (5 nanoseconds,5 nanoseconds,5 nanoseconds)
[ Info: getidx times max(interior,left,right) = (14 nanoseconds,14 nanoseconds,14 nanoseconds)
[ Info: getidx times max(interior,left,right) = (22 nanoseconds,23 nanoseconds,23 nanoseconds)
[ Info: getidx times max(interior,left,right) = (19 nanoseconds,19 nanoseconds,19 nanoseconds)
[ Info: getidx times max(interior,left,right) = (35 nanoseconds,36 nanoseconds,36 nanoseconds)
[ Info: getidx times max(interior,left,right) = (51 nanoseconds,43 nanoseconds,46 nanoseconds)
[ Info: getidx times max(interior,left,right) = (16 nanoseconds,17 nanoseconds,17 nanoseconds)
[ Info: getidx times max(interior,left,right) = (232 nanoseconds,297 nanoseconds,298 nanoseconds)
[ Info: getidx times max(interior,left,right) = (260 nanoseconds,334 nanoseconds,376 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 microseconds, 197 nanoseconds,5 microseconds, 570 nanoseconds,5 microseconds, 584 nanoseconds)
[ Info: getidx times max(interior,left,right) = (107 nanoseconds,156 nanoseconds,152 nanoseconds)
Test Summary:                    | Pass  Total     Time
Scalar Matrix Field Broadcasting |  112    112  5m23.0s
[ Info: getidx times max(interior,left,right) = (11 nanoseconds,11 nanoseconds,11 nanoseconds)
[ Info: getidx times max(interior,left,right) = (93 nanoseconds,96 nanoseconds,120 nanoseconds)
[ Info: getidx times max(interior,left,right) = (48 nanoseconds,48 nanoseconds,48 nanoseconds)
[ Info: getidx times max(interior,left,right) = (70 nanoseconds,77 nanoseconds,77 nanoseconds)
Test Summary:                        | Pass  Total     Time
Non-scalar Matrix Field Broadcasting |   28     28  1m07.7s

this PR

julia> using Revise; include(joinpath("test", "MatrixFields", "matrix_field_broadcasting.jl"))
[ Info: getidx times max(interior,left,right) = (3 nanoseconds,3 nanoseconds,3 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,4 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,4 nanoseconds)
[ Info: getidx times max(interior,left,right) = (4 nanoseconds,4 nanoseconds,4 nanoseconds)
[ Info: getidx times max(interior,left,right) = (7 nanoseconds,7 nanoseconds,7 nanoseconds)
[ Info: getidx times max(interior,left,right) = (5 nanoseconds,5 nanoseconds,5 nanoseconds)
[ Info: getidx times max(interior,left,right) = (14 nanoseconds,15 nanoseconds,14 nanoseconds)
[ Info: getidx times max(interior,left,right) = (22 nanoseconds,23 nanoseconds,23 nanoseconds)
[ Info: getidx times max(interior,left,right) = (19 nanoseconds,19 nanoseconds,19 nanoseconds)
[ Info: getidx times max(interior,left,right) = (35 nanoseconds,36 nanoseconds,36 nanoseconds)
[ Info: getidx times max(interior,left,right) = (15 nanoseconds,16 nanoseconds,16 nanoseconds)
[ Info: getidx times max(interior,left,right) = (15 nanoseconds,16 nanoseconds,16 nanoseconds)
[ Info: getidx times max(interior,left,right) = (57 nanoseconds,82 nanoseconds,93 nanoseconds)
[ Info: getidx times max(interior,left,right) = (134 nanoseconds,163 nanoseconds,182 nanoseconds)
[ Info: getidx times max(interior,left,right) = (723 nanoseconds,784 nanoseconds,829 nanoseconds)
[ Info: getidx times max(interior,left,right) = (97 nanoseconds,129 nanoseconds,130 nanoseconds)
Test Summary:                    | Pass  Total     Time
Scalar Matrix Field Broadcasting |  112    112  5m36.8s
[ Info: getidx times max(interior,left,right) = (11 nanoseconds,11 nanoseconds,11 nanoseconds)
[ Info: getidx times max(interior,left,right) = (95 nanoseconds,110 nanoseconds,99 nanoseconds)
[ Info: getidx times max(interior,left,right) = (48 nanoseconds,49 nanoseconds,49 nanoseconds)
[ Info: getidx times max(interior,left,right) = (70 nanoseconds,78 nanoseconds,78 nanoseconds)
Test Summary:                        | Pass  Total     Time
Non-scalar Matrix Field Broadcasting |   28     28  1m12.2s

So, scalar tests 11, 13, 14, and 15 are significantly faster with this PR.

The flame graph for 15 is a step towards #1871, but there is still some time in eltype, which seems odd.