Add and fix some aqua tests

[charleskawczynski]

This PR adds and fixes some aqua tests. Closes #1964.

[maleadt]

Can you explain why some of these changes are necessary? They only seem to add duplicate definitions with manually-split type signatures, which doesn't seem great.

[charleskawczynski]

Can you explain why some of these changes are necessary? They only seem to add duplicate definitions with manually-split type signatures, which doesn't seem great.

Sure, here's an explanation based on my best understanding. Let's take this fix as an example:

master

LinearAlgebra.triu(A::Union{$SparseMatrixType{T,M}, Transpose{T,<:$SparseMatrixType}, Adjoint{T,<:$SparseMatrixType}}) where {T,M} = 
    $SparseMatrixType( triu(CuSparseMatrixCOO(A), 0) )

fix

LinearAlgebra.triu(A::$SparseMatrixType{T,M}) where {T,M} = 
    $SparseMatrixType( triu(CuSparseMatrixCOO(A), 0) )
LinearAlgebra.triu(A::Union{Transpose{T,<:$SparseMatrixType}, Adjoint{T,<:$SparseMatrixType}}) where {T} = 
    $SparseMatrixType( triu(CuSparseMatrixCOO(A), 0) )

In the main commit, Julia creates methods for each element of the Union. In the case of the second element, the created method is:

LinearAlgebra.triu(A::Transpose{T,<:$SparseMatrixType) where {T,M} = 
    $SparseMatrixType( triu(CuSparseMatrixCOO(A), 0) )

In this method definition, type M is unused and may suffer from JuliaLang/julia#29393. In the fix, the first method was separated into a separate method, but only with the type parameters that are used, so it has no unbound arguments. The second method definition omits the type parameter M, so it too excludes unbound type arguments.

I apologize about the duplication, I tried my best to keep duplication minimal (e.g., creating a shared _unsafe_wrapped_buff from unsafe_wrap). I'm not sure there's alternatives to fix these issues, unfortunately. Perhaps we could ask for JuliaLang/julia#29393 to be reopened, since the issue was closed but technically remains an issue?

[maleadt]

OK thanks, that clears it up. Pushed some additional things, this should be good to go if CI agrees.

[charleskawczynski]

Thanks!