perf(keccak): trim c_aux and skip zero-rotation split witness

[hero78119]

Problem

lookup_keccakf in ceno_zkvm/src/precompiles/lookup_keccakf.rs had redundant witness materialization in a prover hot path:

1. c_aux stored the j=0 prefix stage, which is a direct copy of input state bytes.
2. The RhoPi lane with rotation 0 still used split-rotation witness/range checks.

This increased committed columns and per-round range lookup load without adding new information.

Design Rationale

This change trims only redundant witness encoding while preserving Keccak semantics and existing lookup model (u8 x u8 -> u8):

• Keep the same protocol flow and lookup primitives.
• Remove only derivable witness data:
  • store c_aux only for j=1..4,
  • special-case rot == 0 lane with direct byte equality instead of split witness decomposition.
• Update lookup accounting constants to match the reduced witness/range-check footprint.

Trade-off: slightly more branching in builder/witness generation for reduced witness width and range lookups.

Change Highlights

• ceno_zkvm
  • KeccakWitCols::c_aux reduced from 200 to 160 (drop stored j=0 prefix copy).
  • Theta prefix-XOR constraints updated to derive predecessor from state8[0][i] for j=1.
  • Witness generation updated to mirror new c_aux layout.
  • rotation_witness reduced from 196 to 192.
  • RhoPi rot == 0 lane constrained by direct byte equality (skip split witness/range checks for that lane).
  • RANGE_LOOKUPS_PER_ROUND reduced from 290 to 286.
  • Keccak lookup multiplicity recording kept consistent with the new layout.
  • Commit: 0280726b.

Benchmark / Performance Impact

This PR is performance-oriented (column/lookup reduction). End-to-end benchmark numbers are not included yet in this PR description.

Testing

cargo check -p ceno_zkvm --lib
RUST_MIN_STACK=67108864 cargo test -p ceno_zkvm lookup_keccakf::tests::test_keccakf -- --nocapture

Copilot Reviewer Directive (keep this section)

When Copilot reviews this PR, apply .github/copilot-instructions.md strictly.

[Copilot]

Pull request overview

Performance-focused cleanup of the Keccak-f lookup-based precompile circuit, reducing redundant witness materialization and a small amount of per-round range-check/rotation-witness work while keeping the same lookup-based constraint model.

Changes:

• Shrinks c_aux witness materialization by dropping the j=0 prefix stage and updating Theta prefix-XOR constraints/witness generation accordingly.
• Skips split-rotation witnesses/range checks for the single rot == 0 RhoPi lane by constraining direct byte equality instead.
• Updates per-round accounting constants and refactors Chi byte constraints / multiplicity recording via small helper functions.

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: failure
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24873584569

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: cancelled
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24877666536

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: cancelled
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24877620918

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: cancelled
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24878286573

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: cancelled
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24879312539

[hero78119]

Benchmark result from scroll-tech/ceno-reth-benchmark.

• ceno version: e56d9ba701121c1b68f9e8938553d3725b3fc6ae
• benchmark job: cancelled
• workflow run: https://github.com/scroll-tech/ceno-reth-benchmark/actions/runs/24974514657