Cleanup copy_oftype-like functions in factorizations

[dkarrasch]

Minor cleanup after #40831.

I don't think we need non-mutating lu- and schur-methods with StridedMatrix type in their signature. copy(::StridedMatrix) does the same as copy_similar(::StridedMatrix), so that reduces the number of needed methods.

Closes JuliaDiff/TaylorSeries.jl#295

This now also closes #18970 in a much less verbose way than #41288.

[dkarrasch]

@nanosoldier runtests(["TaylorSeries"], vs = ":master")

[nanosoldier]

Your package evaluation job has completed - no new issues were detected. A full report can be found here.

[dkarrasch]

Pinging @daanhb @andreasnoack

[daanhb]

This looks good to me. It is definitely good to reduce the number of the copy_X functions. For the issue with TaylorSeries, I think a problem might still remain if you supply a symmetric matrix with a different eltype (see my comment in #40831, I had replied there before seeing this)?

[dkarrasch]

I picked up your proposal in the upcoming commit to this PR. I believe we will want get rid of copy_toarray (or, equivalently, Array{S}(A)), since packages that have their own array hierarchy, like CuArray, may want to obtain their own most generic array type, and not necessarily a plain Array.