Optimize prove_f row products

[Kuhai9801]

Summary

• lift q_1 into Zt::CombR once when the conversion is safe
• compute per-row products in the integer ring instead of repeated field conversion
• fall back to the existing field inner product path when conversion is unavailable or checked accumulation overflows
• add focused coverage for direct prove_f usage and field-element integer parsing

Verification

• cargo fmt -p zip-plus --all
• cargo test -p zip-plus --lib phase_prove -- --nocapture
• cargo test -p zip-plus --lib phase_verify -- --nocapture
• cargo check -p zinc-protocol

[albert-garreta]

@Kuhai9801 thanks for the PR! Do you know how much it improves the prover time?

[Kuhai9801]

Thanks @albert-garreta for asking for concrete prover-time numbers here. I opened #201 as a narrower replacement that uses the one-hot structure of Boolean equality tensors in ZipPlus::prove_f and keeps the dense fallback for non-Boolean points.

The PR body includes Criterion results from the existing Zip+ prove benchmarks. On BPoly<63>, the measured speedups are 1.28x for RAA batch 1 at 2^16, 1.36x for RAA batch 5 at 2^16, and 4.59x for IPRS batch 1 at 2^13.