Specialize 5-arg Hermitian-Adjoint multiplication

[jishnub]

This PR computes H::Hermitian * A::Adjoint as (A' * H)'. This often ensures that we take BLAS paths. The 2-arg multiplication was already specialized for StridedMatrix{<:BlasFloat } parents by materializing the Adjoint, but this PR removes this and specializes the 5-arg mul! instead for generic element types. Even if we don't hit BLAS, the change ensures that we iterate in a cache-friendly manner.

However, this PR changes the return type of the 2-arg product from a Matrix{T} to an Adjoint{T, Matrix{T}}

On nightly v"1.13.0-DEV.789"

julia> A = [1 2; 3 4];

julia> Hermitian(A) * A'
2×2 Matrix{Int64}:
  5  11
 10  22

This PR:

julia> Hermitian(A) * A'
2×2 adjoint(::Matrix{Int64}) with eltype Int64:
  5  11
 10  22

In principle, we may copy the result and return a materialized Matrix here.

Some performance improvements:

julia> X = randn(100,100); Y = randn(100,100); Z = copy(X);

julia> @btime Hermitian($X) * $Y';
  34.965 μs (6 allocations: 156.42 KiB) # v"1.13.0-DEV.789"
  23.457 μs (3 allocations: 78.21 KiB) # this PR

julia> @btime mul!($Z', Hermitian($X), $Y', 2.0, 1.0);
  1.311 ms (0 allocations: 0 bytes) # v"1.13.0-DEV.789"
  18.966 μs (0 allocations: 0 bytes) # this PR

julia> XInt, YInt, ZInt = rand(Int,size(X)), rand(Int,size(Y)), rand(Int, size(Z)); # non-BLAS

julia> @btime mul!($ZInt', Hermitian($XInt), $YInt', 2, 1);
  1.031 ms (0 allocations: 0 bytes) # v"1.13.0-DEV.789"
  232.422 μs (0 allocations: 0 bytes) # this PR

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.86%. Comparing base (98a0f15) to head (9129696).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1394      +/-   ##
==========================================
- Coverage   93.87%   93.86%   -0.01%     
==========================================
  Files          34       34              
  Lines       15812    15814       +2     
==========================================
+ Hits        14843    14844       +1     
- Misses        969      970       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.