feat: optimisation of subdivision process

[DavePearce]

Note

Medium Risk
Touches core arithmetic constraint splitting logic; incorrect overflow detection or subdivision could change generated constraints or cause panics, though the new behavior is guarded by explicit overflow checks and a loop-stuck fail-fast.

Overview
Assignment splitting is reworked to reduce unnecessary subdivision. Assignment.chunkUp now iteratively applies a RegisterSplitter, detects which RHS sub-polynomials exceed field.BandWidth via a new determineRhsChunks that returns an overflow register set, and only subdivides those registers (with a stuck-loop guard that panics with debug output).

Experimental path remains but is gated. A USE_NEW_CHUNKUP flag selects the older chunkUp2 path (still present, with determineRhsChunks2 retained under nolint), but defaults to false so the new iterative splitter is the default.

Minor cleanup/robustness. Removes perf-stat logging from lowering/vectorization, and makes debug.PrintSchemas skip printing the concrete schema when absent (HasConcreteSchema()).

^{Written by Cursor Bugbot for commit 9e7f8f5. This will update automatically on new commits. Configure here.}