Summary

Fixes the e2e_fees fee_settings > handles min fee spikes with default padding failure in the merge-train/fairies CI run after PR #23048 was merged (log).

The test captures stableMinFees right after TestContract is deployed (a local peak in feePerL2Gas), then proves two txs (slow), then calls bumpL2Fees(). By the time bumpL2Fees() runs, the chain's min fees have decayed below stableMinFees. A single 20% proving-cost bump only adds ~5% to total feePerL2Gas, which is not enough to climb back above the captured peak — so expect(stableMinFees < bumpedMinFees) fails.

bumpL2Fees now accepts an optional minTarget and bumps proving cost in a loop (each iteration waits for the bump to take effect via the existing retryUntil) until the chain's current min fees exceed the target. Both tests pass their baseline so the comparison is robust to decay. The padded-fee bound is target.mul(1 + DEFAULT_MIN_FEE_PADDING) (= target * 6), so there is plenty of headroom on the upper side.

PR #23048 itself is a pxe tagging refactor and doesn't touch fees — this is a pre-existing flake in the test that surfaced under this run's timing. The same approach was previously prepared in #22583, which was auto-closed only because it went stale.

Notes

• ./bootstrap.sh ci-fast could not be exercised in this container (UID mismatch on ~/.cache blocks the noir bootstrap, no Docker for spartan). The merge-train CI will exercise the change directly.
• Full analysis: https://gist.github.com/AztecBot/5f403a05881febd8fb1b385dddaf323d

ClaudeBox log: https://claudebox.work/s/c40fed098084dd7b?run=2