fix(ci): give AVM check-circuit more CPU/time for heavy txs (canonical)

[AztecBot]

What

Raise the per-command resource budget for avm_check_circuit in yarn-project/end-to-end/bootstrap.sh from CPUS=2 (default) / TIMEOUT=30s to CPUS=4 / TIMEOUT=120s.

Why

The "AVM Circuit Inputs Collection and Check" workflow (avm-check-circuit job) has been failing repeatedly on next. The GitHub job exits 124, which propagates up from a single check-circuit invocation that blows the per-tx timeout while every other input passes in 4–11s. This is a clean wall-clock timeout (exit 124) — not a circuit/assertion error and not OOM (peak ~3.9 GiB inside an 8 GiB container; dmesg showed no kill).

avm_check_circuit_cmds fans out up to 96 bb-avm avm_check_circuit jobs in parallel via parallelize. The heaviest e2e txs build large AVM circuits (~700k rows); trace generation + check-circuit for them does not fit in the long-standing 30s/2-CPU budget (in place since #18747), especially under that parallel CPU contention. The existing code comment already anticipated exactly this ("transactions could need more CPU and MEM than we allocate by default … they might start timing out"). These failures are unrelated to whatever commit happens to be at the head of the failing run — avm_check_circuit is standalone bb-avm reading a dumped .bin.

Observed timeouts span multiple heavy txs, confirming this is not a single-tx fluke:

• e2e_multiple_blobs tx 0x242e67c9… — 34–35s (run 27860137759, and prior run 27805430170)
• e2e_multiple_blobs tx 0x23d0ab72… — 33s (run 28024173324)
• e2e_block_building tx 0x0737b400… — 39s (run 28034181648)

Fix

Both trace generation and check-circuit are multithreaded, so the bottleneck is the 2-CPU cap as much as the 30s clock. Bump to CPUS=4 (which also raises the derived MEM to 16 GiB via MEM=CPUS*4 g) and TIMEOUT=120s for generous headroom on the heaviest txs and on a loaded runner. One-line prefix update plus a refreshed comment. No code/circuit behavior changes; this only adjusts CI execution resources.

Validation note

avm-circuit-inputs.yml triggers only on push to next, the nightly cron, and workflow_dispatch — it does not run on pull requests. So this change cannot be validated by PR CI; it takes effect (and is validated by the next push/nightly run) once merged to next, or via a manual workflow_dispatch on this branch.

Supersedes

This is the canonical consolidation of the duplicate timeout-bump PRs opened by successive failure auto-dispatches. It strictly dominates them (more CPU and a larger global timeout, covering every heavy tx — not just e2e_multiple_blobs):

• fix(ci): bump AVM check-circuit per-tx timeout 30s -> 60s #24208 (global 30s→60s)
• fix(ci): give AVM check-circuit more time for large traces #24193 (global 30s→60s)
• fix(ci): increase AVM check-circuit input timeout #24185 (global 30s→60s via env override)
• fix(ci): extend AVM check-circuit timeout for multiple blobs #24177 (per-dir 2m for e2e_multiple_blobs only — would not cover the e2e_block_building timeout)

Closing those four in favor of this PR.