chore: update reth to 1.11.2

[Qumeric]

Note: this run was for 1.11.0, ideally shall be updated for 1.11.2 too

• Benchmark run
• Baseline

Note that newer rust compiler is likely faster.

Claude's benches Comparison

Document 1 = baseline (main branch), Document 2 = updated (reth-1.11.0 branch). Δ% = (updated - baseline) / baseline.

Summary

Main Cells Used

From 55,890,443 to 56,253,133 (+0.6%)

Stage	Baseline Proof Time (s)	Updated Proof Time (s)	Δ%	Baseline Parallel (s)	Updated Parallel (s)	Δ%	Baseline Parallel/32 (s)	Updated Parallel/32 (s)	Δ%
Total	1,791.20	1,776.26	-0.8%	31.72	34.14	+7.6%	79.42	77.92	-1.9%
reth.prove_stark	737.17	728.35	-1.2%	11.47	12.03	+4.9%	33.42	31.48	-5.8%
leaf	842.92	839.67	-0.4%	7.75	9.82	+26.7%	30.36	31.28	+3.0%
internal.0	155.77	154.60	-0.8%	3.40	3.81	+12.1%	6.55	6.67	+1.8%
internal.1	36.62	35.15	-4.0%	2.56	2.33	-9.0%	2.56	2.33	-9.0%
internal.2	12.33	13.00	+5.4%	2.20	2.62	+19.1%	2.20	2.62	+19.1%
internal.3	4.10	3.10	-24.4%	2.06	2.04	-1.0%	2.06	2.04	-1.0%
internal.4	2.28	1.49	-34.6%	2.28	1.49	-34.6%	2.28	1.49	-34.6%

---

Key takeaways

Wins: internal.4 is dramatically faster (-35% wall time, -66% PCS opening), internal.3 is -24%, and internal.1 is -4%. Total sequential proof time is down ~15s (-0.8%). The generate_perm_trace in reth.prove_stark dropped ~9%.

Regressions: The parallel proof time increased by +7.6% (31.7s → 34.1s), driven primarily by leaf's parallel time jumping +27% (7.75s → 9.82s). internal.0's pcs_opening_time_ms avg is up +5.2% and max jumped from 989ms to 1,404ms. The agg_keygen stark_prove time also crept up (+6.5%).

The internal.4 improvement looks like the most significant structural change — same cell counts but radically less PCS work, suggesting a circuit or constraint change at the top aggregation level.

[jonathanpwang]

I think looks good, but we'll want to wait to merge this when synced with backend @luffykai

[Qumeric]

I am going to update to 1.11.2 as they have high severity bugfix. CC @luffykai