Optimize build_cartesian_ncc_matrix

[tjc0726]

This PR optimizes Product.build_cartesian_ncc_matrix() in dedalus/core/arithmetic.py to achieve speedup for problems with spatially varying coefficients (NCC).

The key insight is to batch-accumulate sparse matrices instead of sequential addition, reducing sorting/deduplication overhead from O(n²) to O(n log n).

In my case

Benchmark: 2D Navier-Stokes with NCC, N=64

• Original: 297.63s per Newton iteration
• Optimized: 27.23s per Newton iteration
• Speedup: 10.93x

Numerical Verification
All fields match to machine precision (relative error < 1e-14):

• u: 3.789e-14
• p: 2.028e-12
• force_x: 6.609e-16

here's my example using monkry patch, one can delete the patch to check the difference

example_ncc_flow_optimized.py