Hoist invariant sum() out of calculate_force_terms inner loop (~16% faster)

[ibell]

Summary

Scroll.calculate_force_terms recomputed sum(_slice) on every iteration of its for CVkey: for theta: double loop, inside the condition if len(error_messages) == sum(_slice). Since _slice = list(range(self.Itheta+1)) is invariant across both loops, this summed ~Itheta integers ~671,000 times per representative scroll run — about 18% of total solve time.

This hoists it to a single n_slice_sum = sum(_slice) before the loops.

Impact

• ~16% faster end-to-end on examples/scroll_compressor.py (16.4s → 13.7s), interpreted/default build — no compilation, no API change.
• Bit-identical results: volumetric efficiency (91.95998551975654 %), adiabatic efficiency (74.61003492585036 %), mass flow (55.175673362688805 g/s), and the force/moment outputs all match to full precision before and after.

How it was found

Native stack sampling of the unprofiled solve showed builtins.sum at ~18% of solve time; cProfile masks this because its per-call overhead inflates the millions of cheap delegation calls instead. cProfile's call graph (print_callers) then pinned the caller to calculate_force_terms.

Note (not changed here)

The condition len(error_messages) == sum(_slice) looks like it may have been intended as len(...) == len(_slice) (i.e. "every force evaluation for this CV failed, so print the unique errors"). This PR preserves the existing behavior exactly and only removes the redundant recomputation; the suspected logic question is left for a separate decision.

Test Plan

• examples/scroll_compressor.py runs and all reported quantities are bit-identical to master
• ~16% wall-clock reduction confirmed across repeated runs

🤖 Generated with Claude Code