Fix KLUFactorization for AbstractSparseMatrixCSC subtypes

Fixes #737 by handling the case where the cached KLU factorization is
uninitialized (PREALLOCATED_KLU) or has a dimension mismatch with the
input matrix.

The issue occurred when using AbstractSparseMatrixCSC subtypes because:
1. The cacheval would be set to PREALLOCATED_KLU (a 0x0 matrix)
2. pattern_changed would return false for this case
3. klu! would be called with mismatched dimensions, causing an error

The fix checks if the cacheval is uninitialized or has a size mismatch,
and creates a new factorization in those cases instead of trying to
reuse the incompatible cached value.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

Author: claude

ext/LinearSolveSparseArraysExt.jl

@@ -259,6 +259,13 @@ function SciMLBase.solve!(cache::LinearSolve.LinearCache, alg::KLUFactorization;
                     SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
                         nonzeros(A)),
                     check = false)
+            elseif cacheval === nothing || cacheval === PREALLOCATED_KLU ||
+                   length(nonzeros(A)) != length(cacheval.nzval)
+                # Create new factorization if cacheval is uninitialized or size mismatch
+                fact = KLU.klu(
+                    SparseMatrixCSC(size(A)..., getcolptr(A), rowvals(A),
+                        nonzeros(A)),
+                    check = false)
             else
                 fact = KLU.klu!(cacheval, nonzeros(A), check = false)
             end