test: close root-cause race in warpL2TimeAtLeastBy timestamp check

[AztecBot]

Why

The merge-train/spartan-v5 train PR (#24272) was dequeued from the merge queue. Investigating the two merge-queue runs showed two distinct causes:

• First dequeue (merge-queue run at 18:34 UTC, before test: retry composed cheat code timestamp race #24279 landed): a genuine test failure — e2e_cheat_codes › warpL2TimeAtLeastBy with sub-slot duration auto-adjusts to next slot threw warpL2TimeAtLeastTo: target timestamp <T> is not in the future (current L1 timestamp is <T+11>). This is the composed cheat-code timestamp race.
• Most recent dequeue (merge-queue run at 01:40 UTC): infrastructure, not a test failure — the x8-full grind's build instance went unreachable (SSM command … did not succeed … status=Failed, statusDetails=Undeliverable, responseCode=-1) after ~76 min; the other 7 grinds passed with #24279 present.

#24279 mitigated the race at the test level with a 5-attempt retry helper. This PR removes the race at its root so the retry is no longer load-bearing.

The race

warpL2TimeAtLeastBy reads the L1 timestamp to derive target = max(L2, L1) + duration, then delegated to warpL2TimeAtLeastTo, which re-read L1 for its future-check guard. A live sequencer can warp L1 forward by a full slot between those two reads, so for a small sub-slot duration the freshly-read L1 overtakes target and the guard throws is not in the future — even though the downstream slot/mine logic uses fresh rollup state and would have advanced time correctly.

The fix

Funnel both public methods through a private #warpL2TimeAtLeastTo(node, targetBigInt, currentTimestamp) that takes the caller's already-observed L1 timestamp. warpL2TimeAtLeastBy now reuses the single L1 read its target was derived from, so target = base + duration (with base ≥ currentL1 and duration > 0) is always strictly in the future — the throw window is gone. The public warpL2TimeAtLeastTo still reads L1 itself for direct callers, preserving the warpL2TimeAtLeastTo with past timestamp throws behavior. Public signatures are unchanged.

Testing

• Static: change is isolated to cheat_codes.ts; all external callers use the unchanged public methods. A full local ./bootstrap.sh ci was not runnable in this environment (it orchestrates remote EC2 grind instances and the composed e2e_cheat_codes test needs a live docker-compose network), so verification runs through the train's merge-queue CI.

---

Created by claudebox · group: slackbot

[AztecBot]

Automatically closing this stale claudebox draft PR (no updates for 5+ days). Re-open if still needed.