Add allocation tests for caching interface

[ChrisRackauckas-Claude]

Summary

This PR adds comprehensive allocation tests for the caching interface in LinearSolve.jl to ensure that the caching mechanism doesn't allocate unnecessarily when reusing factorizations, as documented in https://docs.sciml.ai/LinearSolve/stable/tutorials/caching_interface/

Changes

• Added test/nopre/caching_allocation_tests.jl with comprehensive allocation tests
• Updated test/nopre/Project.toml to include AllocCheck.jl for allocation testing
• Added the new test file to test/runtests.jl in the NoPre test group

Test Coverage

The tests verify allocation-free behavior for:

Dense Factorizations

• LUFactorization
• QRFactorization
• CholeskyFactorization
• SVDFactorization
• BunchKaufmanFactorization
• NormalCholeskyFactorization
• DiagonalFactorization

Sparse Factorizations

• KLUFactorization
• UMFPACKFactorization
• CHOLMODFactorization

Iterative Solvers

• SimpleGMRES
• KrylovJL_GMRES
• KrylovJL_CG
• KrylovJL_BICGSTAB

Key Test Scenarios

1. Allocation-free solving with different b vectors - Verifies that changing only the right-hand side vector and resolving doesn't allocate after initial factorization
2. Matrix change triggers refactorization - Ensures that changing the matrix properly triggers refactorization (allocations expected)
3. Non-square matrix handling - Tests overdetermined and underdetermined systems with appropriate residual checks
4. Performance comparison - Compares caching vs non-caching approaches to verify the benefit

Implementation Notes

• Uses @check_allocs from AllocCheck.jl to verify zero allocations
• Some algorithms may still allocate in certain Julia versions (marked as @test_broken where appropriate)
• Tests follow the caching interface pattern: init → solve\! → modify cache.b → solve\! again

Test Results

All tests pass successfully, with some expected @test_broken for algorithms that haven't been optimized for zero allocations yet.

🤖 Generated with Claude Code