Remove LUFactorization method for AbstractSciMLOperator

The LUFactorization method for AbstractSciMLOperator was causing issues
with the default algorithm selection. This removes it while keeping
support for KLU and UMFPACK factorizations with operators that have
concretization support.

Fixes #848

Author: claude

ext/LinearSolveSparseArraysExt.jl

@@ -300,18 +300,6 @@ function LinearSolve.init_cacheval(
     end
 end
 
-function LinearSolve.init_cacheval(
-        alg::LUFactorization, A::AbstractSciMLOperator, b, u, Pl, Pr,
-        maxiters::Int, abstol, reltol,
-        verbose::Union{LinearVerbosity, Bool}, assumptions::OperatorAssumptions)
-    if has_concretization(A)
-        return LinearSolve.init_cacheval(alg, convert(AbstractMatrix, A), b, u, Pl, Pr,
-            maxiters, abstol, reltol, verbose, assumptions)
-    else
-        nothing
-    end
-end
-
 function LinearSolve.init_cacheval(
         alg::CHOLMODFactorization, A::AbstractSciMLOperator, b, u, Pl, Pr,
         maxiters::Int, abstol, reltol,

test/basictests.jl

@@ -584,11 +584,6 @@ end
         sol_operator = solve(prob_operator, UMFPACKFactorization())
         @test sol_matrix.u ≈ sol_operator.u
 
-        # Test LU with operator
-        sol_matrix = solve(prob_matrix, LUFactorization())
-        sol_operator = solve(prob_operator, LUFactorization())
-        @test sol_matrix.u ≈ sol_operator.u
-
         # Test WOperator with sparse Jacobian
         n_w = 8
         M = sparse(I(n_w) * 1.0)
@@ -612,11 +607,6 @@ end
         sol_woperator = solve(prob_woperator, UMFPACKFactorization())
         sol_wmatrix = solve(prob_wmatrix, UMFPACKFactorization())
         @test sol_woperator.u ≈ sol_wmatrix.u
-
-        # Test LU with WOperator
-        sol_woperator = solve(prob_woperator, LUFactorization())
-        sol_wmatrix = solve(prob_wmatrix, LUFactorization())
-        @test sol_woperator.u ≈ sol_wmatrix.u
     end
 
     @testset "Solve Function" begin