Consolidate more benchmark code, make ff optimize more robust

[scal444]

Consolidated CPU count with fallback for older python (< 3.13).

Added pre-run checks for MMFF/UFF forcefield compatibility so we don't choke on inputs with a few bad molecules.

[greptile-apps]

Greptile Summary

Consolidates benchmark CPU availability detection and makes force-field optimization more resilient.

• Adds a cross-version, affinity-aware CPU-count helper and adopts it across benchmark scripts.
• Adds optional buffered molecule loading to compensate for preparation and force-field filtering losses.
• Filters molecules lacking MMFF or UFF parameters before conformer generation and optimization.
• Separates conformer-generation process count from RDKit optimizer thread count.
• Adds focused tests for candidate buffering and Python 3.12 CPU-count fallback behavior.

Confidence Score: 4/5

The PR is safe to merge, with a non-blocking concern about excessive conformer-generation workers on high-core-count machines.

Loader and force-field behavior remain compatible, while the new force-field benchmark default starts one conformer process per available CPU and can impose avoidable resource overhead.

Files Needing Attention: benchmarks/ff_optimize_bench.py

Important Files Changed

Filename	Overview
benchmarks/bench_utils/loaders.py	Adds an opt-in 10% candidate reserve while preserving existing loader bounds and sampling behavior by default.
benchmarks/bench_utils/molprep.py	Adds a safe Linux-compatible CPU availability helper with fallbacks across supported Python versions.
benchmarks/ff_optimize_bench.py	Buffers and filters force-field inputs and adds a dedicated conformer process setting, but its automatic default uses every available CPU rather than the repository's half-CPU convention.
benchmarks/tests/test_loaders.py	Covers minimum buffer sizing, buffered SMILES loading, and the pre-Python-3.14 CPU-count fallback.
benchmarks/conformer_rmsd_bench.py	Replaces the multiprocessing CPU count with the shared affinity-aware helper.
benchmarks/tfd_bench.py	Uses the shared CPU-count helper while retaining the existing half-CPU worker default.
benchmarks/tfd_prepare_mols.py	Consolidates default worker calculation through the shared CPU-count helper.

_{Reviews (1): Last reviewed commit: "Consolidate more benchmark code, make ff..." | Re-trigger Greptile}

[evasnow1992]

Thank you for the consolidation. Changes are clean.