Work with Distributions.jl and PDMats.jl

[Wu-Chenyang]

Currently, the Kronecker can't work with Distributions.jl and PDMats.jl. The following simple code fails.

julia> using Kronecker, Distributions, LinearAlgebra
julia> MvNormal(ones(4), I(2)⊗I(2))
ERROR: MethodError: no method matching PDMats.PDMat(::Kronecker.KroneckerProduct{Bool, Diagonal{…}, Diagonal{…}}, ::CholeskyKronecker{Cholesky{…}, Cholesky{…}})
The type `PDMats.PDMat` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  PDMats.PDMat(::AbstractMatrix, ::Cholesky{T, S}) where {T, S}
   @ PDMats ~/.julia/packages/PDMats/cAM9h/src/pdmat.jl:11
  PDMats.PDMat(::AbstractMatrix)
   @ PDMats ~/.julia/packages/PDMats/cAM9h/src/pdmat.jl:19
  PDMats.PDMat(::Cholesky)
   @ PDMats ~/.julia/packages/PDMats/cAM9h/src/pdmat.jl:20

Is there a workaround for this? It seems requiring CholeskyKronecker to subtype Cholesky.

[MichielStock]

The issue is that Distributions seems to use PDMmat for the covariance matrix. It would seem useful to be able to use the Matrix normal then.

To avoid type piracy, I think the best option is to copy this issue to PDMats.jl. I'll open an issue there.