Co-authored-by: Copilot <copilot@github.com>

…_batched`

Co-authored-by: Copilot <copilot@github.com>

# Conflicts:
#	crates/whir/src/open.rs

Adopt main's always-IV slice hashing (length absorbed into the IV for
domain separation) while keeping whir-split-eq's existing zero-skip
optimization at the DFT layer (commit_with_prefix_len /
reorder_and_dft_with_prefix_len). The Merkle layer keeps the simpler
merkle_commit(matrix, n_cols) API and hashes the full DFT matrix,
since the DFT already cut trailing zeros.

Concretely:
- precompute_zero_suffix_state + hash_rtl_iter_with_initial_state added
  to the sponge module (alongside the existing hash_iter).
- whir/src/merkle.rs first_digest_layer switched to RTL absorption
  seeded with [base_width, 0, ..., 0], via the new
  vertically_packed_row_rtl matrix helper (n_zero_chunks = 0).
- backend merkle_verify and fiat-shamir verifier swap hash_iter for
  hash_slice so the verifier reproduces the prover's IV-based digest.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Adapts main's 'Custom thread pool (#239)' (rayon -> parallel crate) to the
whir-split-eq refactor. This branch's SVO / LSB-first-fold WHIR prover supersedes
main's WHIR statement-combining: kept whir-split-eq's unpacked Vec<EF> weights,
compute_eval_eq_base_batched, LSB-cols FFT-prep layout, and svo.rs, dropping main's
superseded combine_statement / compute_eval_eq_base_packed_batched. Converted all of
whir-split-eq's rayon (svo.rs, open.rs) to the pool, and made build_accumulators
sequential across groups since the pool forbids the nested dispatch rayon allowed
(each group still parallelizes internally). Adapted lsb_fold to main's reworked
fold_multilinear_at_bit API.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Reconcile main's arena-allocator migration (#247, "remove global_allocator
in zk-alloc") with the WHIR split-eq refactor on this branch.

Conflicts resolved:
- quotient_gkr/layers.rs: materialise_in_full keeps the branch's pub(super)
  visibility and adopts main's ArenaVec return type.
- whir/utils.rs: keep the split-eq LSB-cols FFT-prep layout but write into an
  ArenaVec via par_fill (main's allocator) instead of par_map_collect's Vec;
  drop the now-unused log2_strict_usize import.
- whir/open.rs: keep the branch's final_round / open_merkle_tree_at_challenges
  refactor and SVO sumcheck; drop main's combine_statement (deleted on this
  branch). Propagate main's ArenaVec migration through the new SVO helpers
  (SumcheckSingle.weights, lsb_fold, fold_by_tensor, build_post_svo_weights)
  to avoid Vec<->ArenaVec copies in the prover hot path.

Auto-merges verified coherent: eq_mle.rs composes the branch's
compute_eval_eq_base_batched rename/unpacked-output with main's ArenaVec
buffers; commit.rs changes are in disjoint methods.

cargo fmt / clippy clean; cargo testall (release) green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>