Uff python bindings

[scal444]

No description provided.

[greptile-apps]

Greptile Summary

Adds GPU-accelerated UFF force-field optimization as a new Python binding (UFFOptimizeMoleculesConfs), following the established MMFF pattern. Common BFGS helpers are extracted into bfgs_common.cpp/h and shared between the MMFF and UFF backends.

Confidence Score: 5/5

Safe to merge; all findings are P2 style and performance suggestions with no correctness impact.

The implementation closely mirrors the existing MMFF pattern. Three P2 findings: weak stub types, an unchecked embedding return in a test helper, and a missed FF-params caching opportunity. None affect runtime correctness or data integrity.

No files require special attention for merge safety.

Important Files Changed

Filename	Overview
src/minimizer/bfgs_uff.cpp	New UFF BFGS minimizer; missing per-molecule FF-param caching present in the MMFF counterpart
nvmolkit/uffOptimization.py	New Python wrapper for UFF optimization; consistent scalar-or-sequence normalisation matching MMFF pattern
nvmolkit/uffOptimization.cpp	New Boost.Python bindings; clean extraction helpers with null-pointer guard
nvmolkit/tests/test_uff_optimization.py	New UFF test suite; make_fragmented_mol does not check EmbedMultipleConfs return value
nvmolkit/_uffOptimization.pyi	New C++ binding stub; vdwThresholds and ignoreInterfragInteractions typed as Any
src/minimizer/bfgs_common.cpp	Shared BFGS helpers extracted from bfgs_mmff; energies buffer sizing corrected vs. original
src/minimizer/bfgs_common.h	New shared header for ThreadLocalBuffers, ConformerInfo, and BatchExecutionContext
src/minimizer/bfgs_uff.h	New UFF minimizer header; clean signature in dedicated UFF namespace
src/minimizer/bfgs_mmff.cpp	Refactored to use shared bfgs_common helpers; logic and caching pattern unchanged
nvmolkit/CMakeLists.txt	Added _uffOptimization build target following existing _mmffOptimization pattern
nvmolkit/init.py	Minor docstring update noting UFF support
src/minimizer/CMakeLists.txt	Added bfgs_uff.cpp and bfgs_common.cpp source files

_{Reviews (1): Last reviewed commit: "Consolidate things" | Re-trigger Greptile}

[greptile-apps]

Weak Any types reduce stub value

All three non-trivial parameters are typed as Any, eliminating IDE autocompletion and type-checker coverage for callers of the internal binding. The C++ layer already enforces concrete list types, so the stub can reflect them.

Suggested change

	vdwThresholds: Any = ...,
	ignoreInterfragInteractions: Any = ...,
	hardwareOptions: Any = ...,
	vdwThresholds: List[float] = ...,
	ignoreInterfragInteractions: List[bool] = ...,
	hardwareOptions: Any = ...,

[greptile-apps]

EmbedMultipleConfs return value not checked

EmbedMultipleConfs returns a list of assigned conformer IDs (empty on failure). If embedding fails, the unchecked mol.GetConformer() on the next line raises ValueError: No conformers present, making the test failure hard to diagnose.

Suggested change

	rdDistGeom.EmbedMultipleConfs(mol, numConfs=1, params=params)
	conf = mol.GetConformer()
	confs_embedded = rdDistGeom.EmbedMultipleConfs(mol, numConfs=1, params=params)
	if not confs_embedded:
	raise RuntimeError("EmbedMultipleConfs failed for make_fragmented_mol(); check SMILES or random seed")
	conf = mol.GetConformer()

[greptile-apps]

No per-molecule caching of FF params, unlike the MMFF counterpart

constructForcefieldContribs is called once per conformer. The MMFF implementation (bfgs_mmff.cpp) uses a moleculeCache keyed on ROMol* to call this only once per unique molecule per batch, since bond/angle/torsion/vdW topology parameters are conformer-independent. For batches containing molecules with many conformers, UFF re-derives the full force-field topology redundantly on every iteration.