Jishnu Bhattacharya

This PR is primarily about faster compilation. The difference mainly arises from the fact that `copyto!` is specialized for `Array`s to essentially be reduced to a foreigncall to `jl_genericmemory_copyto`. On...

The docstring of `copy_similar` currently says that the type of the output corresponds to `similar(A, T, size(A))` in general (although I'm unsure why the three-term `similar` is specifically necessary). If...

I don't think we should call `copyto!` here with two `Hermitian` matrices, but we should call `copytrito!` directly on the parents (if not `copyto!`). This is because we'll unnecessarily compile...

I've timed the `copytrito!` and the `copyto_similar` implementations, and the performance of `eigen` seems roughly comparable for the two: With ```julia eigencopy_oftype(A::Hermitian, S) = Hermitian(copytrito!(similar(parent(A), S, size(A)), A.data, A.uplo), sym_uplo(A.uplo))...

Yeah, that's one potential catch with using `copytrito!`, where one half of the matrix is left untouched (and potentially uninitialized). In this case, the issue seems to be that `adjoint!`...

The specific failing test should be fixed by https://github.com/JuliaLang/julia/pull/54491

The failing `BunchKaufman` test should be fixed by https://github.com/JuliaLang/julia/pull/54509

The failing `hessenberg` test should be fixed by https://github.com/JuliaLang/julia/pull/54518 (I think)

Indeed, I am happy to close this, but why was the other one stalled?

> This should be on by default (for new users Why though? This is a perfectly reasonable approach, although it's not composable. New users probably aren't writing composable packages anyway....