Divan bench for whir

[zkfriendly]

Summary

The current Divan benchmarks only cover isolated pieces of the protocol (sumcheck, expand_from_coeff/NTT). There is no end-to-end WHIR benchmark in the Divan/CodSpeed setup. E2E WHIR timing does exist via the CLI in src/bin/benchmark.rs, but that tool is not a Divan bench, so it is not run in CI/CodSpeed.

This PR adds a whir Divan bench target with two E2E prover benchmarks on Basefield<Field64_3>:

• whir_ldt — WHIR as a low-degree test, with no linear forms or evaluation claims (commit + prove only).
• whir_pcs — WHIR as a polynomial commitment scheme, with 2 multilinear evaluation openings, following the PCS path in benchmark.rs.

Both sweep vector sizes from 2^16 to 2^21 using fixed protocol parameters (128-bit security, 20 PoW bits, folding factor 4, Blake3). Setup work (vector construction, PCS evaluation computation) stays in with_inputs; the timed section runs the full WHIR prover flow through a shared run_whir helper (domain separator, commit, prove).

Test plan

• cargo fmt --all --check
• cargo clippy --workspace --all-targets --all-features -- -Dwarnings
• cargo test --release --all-features --all-targets (includes compiling/running the new whir bench)
• Confirm CodSpeed picks up the new whir bench on merge (cargo codspeed build / cargo codspeed run)

[codspeed-hq]

Merging this PR will not alter performance

✅ 10 untouched benchmarks
🆕 8 new benchmarks
⏩ 22 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	Simulation	whir_ldt[1048576]	N/A	4.2 s	N/A
🆕	Simulation	whir_ldt[2097152]	N/A	8.6 s	N/A
🆕	Simulation	whir_ldt[262144]	N/A	1.5 s	N/A
🆕	Simulation	whir_ldt[65536]	N/A	882.1 ms	N/A
🆕	Simulation	whir_pcs[1048576]	N/A	4.9 s	N/A
🆕	Simulation	whir_pcs[2097152]	N/A	8.9 s	N/A
🆕	Simulation	whir_pcs[262144]	N/A	1.1 s	N/A
🆕	Simulation	whir_pcs[65536]	N/A	530.5 ms	N/A

---

_{Comparing zkfr/divan-bench-for-whir (0a71a91) with main (10aa7d0)}

Footnotes

1. 22 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩