perf: speed up kernel checking of ring and field certificates (#15944)

This PR speeds up kernel checking of ring and field certificates
produced by `grind` and `Sym.simp`.

Specialize monomial equality by projecting power fields directly, and
inline `Bool.rec` in polynomial equality and one-pass addition checks.
Add recursor-based inverse cancellation and specialized
rational-polynomial equality for field certificates, with equivalence
proofs, differential tests, and a scalable benchmark.

Kernel instruction counts decrease by 2.33% across 12 existing
workloads, including 2.49% for `groebner_hard` and 6.57% for
`sym_arith_norm_field`. The new inverse-cancellation stress benchmark
uses 68.21% fewer kernel instructions. These figures use three-run
medians, subtracting runs with `debug.skipKernelTC` enabled to isolate
kernel checking costs.