Bug fixes, performance improvements, and test coverage

[flaport]

Summary

• Fix C++ memory safety issues: add null checks after klu_factor, use RAII with std::unique_ptr for automatic cleanup, and check Common.status after KLU operations
• Fix incorrect error messages (n_lhs -> n_nz mismatch) and add negative index validation
• Add error handling in JVP functions when trying to differentiate w.r.t. sparse indices Ai/Aj
• Improve dot operation performance by ~4x via loop reordering for better cache locality
• Add klujax.dot to the parametrized test suite (previously only solve was tested)
• Fix test matrices to be invertible by adding strong diagonal (prevents singular matrix errors)
• Update pre-commit hooks and switch from Makefile to justfile

Test plan

• All 36 tests pass (make test / just test)
• Benchmarked dot operation: ~4.3x speedup (2159ms -> 504ms mean per iteration)
• Verified singular matrix detection works correctly

🤖 Generated with Claude Code