Method resolution issue for rdiv!

[palday]

The change in JuliaLang/julia#38168 causes MixedModels to fail on Julia master with a method resolution error for rdiv!:

  MethodError: no method matching rdiv!(::BlockedSparse{Float64, 1, 2}, ::UpperTriangular{Float64, Adjoint{Float64, UniformBlo
ckDiagonal{Float64}}})                                         
  Closest candidates are:
    rdiv!(::StridedMatrix{T} where T, ::UpperTriangular{var"#s814", var"#s813"} where var"#s813"<:Adjoint where var"#s814") at  ~/julia/usr/share/julia/stdlib/v1.6/LinearAlgebra/src/triangular.jl:1430
    rdiv!(::StridedMatrix{T} where T, ::UpperTriangular) at ~/julia/usr/share/julia/stdlib/v1.6/LinearAlgebra
/src/triangular.jl:1327
    rdiv!(::AbstractMatrix{T} where T, ::Transpose{var"#s826", var"#s825"} where var"#s825"<:LU where var"#s826") at ~/julia/usr/share/julia/stdlib/v1.6/LinearAlgebra/src/lu.jl:689

Tests pass again after reverting this change in a local build.

BlockedSparse is a wrapper around SparseMatrixCSC with a particular sparsity pattern. Note that this does not fall back to the generic multiplication method -- the change in order means that the second argument doesn't match the second argument in the last of the closest candidates.

[dkarrasch]

JuliaLang/julia#38168 requires package maintainers to follow with swapping the wrappers. As I suspected in JuliaLang/julia#38168, I guess we should have done some kind of package evaluation before merging. Related error messages may look different though, depending on the specific context.

[palday]

If package evaluation passes, I'm happy to close this issue.