Make inverting a PDSparseMat throw an error

Alter the inv() method for PDSparseMat to fall through to the inv() method of its underlying matrix.  If that is a SparseMatrixCSC, an error will be thrown asking the user to convert to a dense matrix first (the default behavior in LinearAlgebra).

Author: ElOceanografo

src/pdsparsemat.jl

@@ -51,7 +51,7 @@ end
 
 ### Algebra
 
-Base.inv(a::PDSparseMat{T}) where {T<:Real} = PDMat( a\eye(T,a.dim) )
+Base.inv(a::PDSparseMat{T}) where {T<:Real} = PDMat(inv(a.mat))
 LinearAlgebra.logdet(a::PDSparseMat) = logdet(a.chol)
 
 ### whiten and unwhiten