Benchmarkoor: Baseline (before fixes) vs After fixes

Baseline run: 1779064698_0ecabe5f_erigon-bal-full — before any of our fixes; default cpuset_count: 6 config; default lazy KZG init.

After run: 1779109056_a101d4ed_erigon-bal-full — with the BAL tx.Apply read fix, KZG async warmup at backend.New(), and cpuset_count: 6

Image: erigon-local:traced for both. 17 tests, 2 messages each (engine_forkchoiceUpdatedV3 + engine_newPayloadV5). All values are engine_newPayloadV5.last MGas/s. Sorted by speedup desc.

Where the speedup comes from

• test_point_evaluation jumps 5.45× (43.7 → 238.5 MGas/s) primarily from the KZG async warmup in backend.New() — the very first block no longer pays the ~1.2s InitKZGCtx() cost on a worker goroutine.
• The remaining ~2.0–4.7× wins on test_mod, test_p256verify, test_ecrecover, test_blake2f_benchmark, test_mod_arithmetic come from the BAL tx.Apply read fix (PR execution: optimise parallel exec with BALs for same-sender conflicts in precompile benchmarks #21240's same-sender retry storm fix only landed because the BAL bytes are now visible to the validation overlay)