improve transition-metal performance

[corinwagen]

• Add transition-metal conformer support via a new preset, automatic TM move budget, ligand/haptic rotation moves, and TM seed augmentation.
• Centralize geometry constraints and apply metal-shell/pose constraints consistently across MMFF, UFF, and low-mode workflows.
• Preserve protected input/pose seeds through pooling, dedupe, and final selection, including fallback behavior for unminimizable metal inputs.
• Expand docs, runtime tuning, stats, tests, and .gitignore coverage for TM workflows and constrained generation.

Benchmarks seem decent:

• 16tmconf543 / 16 TM complexes: steamroll transition_metal completed 16/16 with median low-5 kcal best-RMSD 1.35 Å, low-5 coverage <=1 Å of 0.36, and low-10 coverage <=1.5 Å of 0.42 in ~26 s. This
  is a large improvement over the prior steamroll baseline (3.68 A, zero coverage).
• TMCONF40: steamroll transition_metal completed 36/40 with median low-5 kcal best-RMSD 1.56 Å, low-5 coverage <=1 Å of 0.26, and low-10 coverage <=1.5 Å of 0.46 in ~44 s. Remaining failures are
  conversion/subprocess failures, not ordinary generation failures.

This isn't yet amazing for transition metals, but it should be less bad.