Optimization: Prune zero weight grid points in horton3 adapter

[Ao-chuba]

Hello , @tovrstra @FarnazH
While exploring src/denspart/adapters/horton3.py, I noticed a TODO comment on line 122 suggesting the removal of grid points with zero weight.
Analysis: I ran an analysis using the examples/horton3 test case (water molecule) to quantify the number of negligible points.

• Total Grid Points: 87,300
• Negligible Points (< 1e-10): ~4,961
• Impact: Approximately 5.7% of the generated grid points effectively contribute nothing to the integration but consume computational resources during density and property evaluations. For larger systems or different grid settings, this percentage could be higher.

Solution: I propose implementing a filter in _setup_grid to remove points with weights below a threshold (e.g., 1e-15) immediately after grid generation. This will reduce the workload for all subsequent tensor operations.

[tovrstra]

Sounds like a good addition. Thanks for picking this up!

[PaulWAyers]

I commented on the pull request. I'm fine with this, but I'm not sure but what it might belong in the grid package instead of in denspart (using grid). It would make sense, to me, to have a threshold in grid pruning out points that have low integration weights and/or points that are very far from the nucleus (which typically, but not always, have small weights).

[Ao-chuba]

@PaulWAyers
You're right ,handling this in grid would be cleaner for the whole ecosystem. I went with this approach mainly because it was simpler and directly addressed the existing TODO in this file. It gets the job done for denspart right now without requiring changes to external dependencies.