perf(engine): reduce proof worker count for small blocks

[gakonst]

Summary

Reduce proof worker pool size for blocks with ≤30 transactions to cut per-block spawn overhead.

Motivation

ProofWorkerHandle::new() dominates spawn_payload_processor cost: ~1.5ms of ~2.5ms total per block. On a 31-core machine it spawns 62 workers (31 storage + 31 account), each opening an MDBX read txn + creating trie cursors.

Profiling shows each worker uses only 3.9μs CPU per block — 0.0005% utilization. They are almost entirely idle for blocks under 30 txs.

We cannot skip SRT entirely (PR #22129 tried, regressed +34%) because PreservedSparseTrie is reused block-to-block via state_root anchoring. Switching to Parallel forces full trie recomputation.

Changes

• Save transaction_count from ExecutionEnv before it is moved
• Externalize storage_worker_count and account_worker_count as params to ProofWorkerHandle::new() instead of reading from runtime
• When tx count ≤30, cap workers at min(pool_size, 16) — preserves SRT cache chain while reducing spawn overhead

Expected: MDBX read txns per block 62→16 for small blocks, spawn overhead 1.5ms→0.4ms.

Bench

Results < 30

<30 txns
         Baseline   Feature   Change  
Mean      16.08ms   15.68ms   -2.51%
StdDev     6.13ms    6.10ms
P50       18.55ms   17.93ms   -3.38% 
P90       20.95ms   19.87ms   -5.16% 
P99       28.77ms   29.38ms   +2.15%
Mgas/s     638.86    655.30   +2.57%

normal:

Baseline   Feature   Change   
Mean      32.85ms   32.47ms   -1.15% 
StdDev    16.97ms   16.44ms
P50       27.94ms   27.84ms   -0.36%
P90       53.93ms    53.5ms    -0.8% 
P99       91.54ms   91.46ms   -0.08% 
Mgas/s      928.8    939.59   +1.16% 

Profiling data: Samply baseline | Samply feature

[github-actions]

⚠️ Changelog not found.

A changelog entry is required before merging. We've generated a suggested changelog based on your changes:

Preview

---
reth-engine-tree: patch
reth-trie-parallel: patch
---

Added adaptive proof worker pool sizing that halves worker count for blocks with 30 or fewer transactions to reduce idle overhead when fewer state changes are generated.

Add changelog to commit this to your branch.

[DaniPopes]

overhead is reduced with rayon, can we rebench?

[yongkangc]

@DaniPopes thanks for note, rebenching