fix(test-benchmark): fill benchmark blocks fully under EIP-8037 gas accounting (repricing)

[jochem-brouwer]

Description

Benchmarkoor runs of the repricing suite showed underfilled blocks (e.g. 82.9% / 89.6% of the block gas limit at 100–300M). Two gas-model gaps, both introduced by the new gas accounting:

1. Value-bearing CALL/CALLCODE overcharged by the stipend (blocks 82.9% full). The model charged the full CALL_VALUE (10,300) but never credited the 2,300 stipend the callee returns unconsumed. Adds an opt-in stipend_returned opcode metadata flag that nets the stipend out of the modeled cost, plus a matching IteratingBytecode gas-limit reserve: the frame must still momentarily hold the gross charge on the last loop iteration, so each transaction's gas limit (and the block packing budget) includes one stipend of headroom. Used by test_account_access for value_sent > 0.

2. test_ether_transfers_onchain_receivers packed by the sum of gas dimensions (blocks 89.6% full for diff_to_nonexistent). Under EIP-8037 the block header gas is the max of the execution and state dimensions, not their sum. Transfers to nonexistent accounts pay NEW_ACCOUNT (183,600) state gas per transaction, so packing by the sum left ~10% of the binding dimension unused. The transaction count is now derived from the binding dimension while each transaction's gas limit still covers both. All other parametrizations produce identical fixtures.

Verified against the EELS spec: a CALL-with-value loop fills a block to the exact modeled gas unit, and the old packing formula reproduces the exact underfilled header gas observed on besu (89,596,800 at 100M → 99,878,400 after the fix).

Note: PR + description mostly done with LLM.

TODO: (this is why PR is draft) verify that this does not introduce any fill bugs on state-actor

Related Issues or PRs

N/A.

Checklist

• Ran fast static checks to avoid CI fails, see Code Standards & Verifying Changes: just static
• PR title has the form <type>(<area>): <title>, where <type> and <area> come from an appropriate C-<type>, respectively A-<area>, label. The title should match the target squash commit message.

Cute Animal Picture

Meerkat is an Afrikaans word (believed?) to be from "mier" (ant) and "kat" (cat) 😄 👍

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.49%. Comparing base (0da4387) to head (b39d963).

Additional details and impacted files

@@               Coverage Diff                @@
##           forks/amsterdam    #3269   +/-   ##
================================================
  Coverage            93.49%   93.49%           
================================================
  Files                  625      625           
  Lines                37032    37032           
  Branches              3385     3385           
================================================
  Hits                 34623    34623           
  Misses                1653     1653           
  Partials               756      756           

Flag	Coverage Δ
unittests	93.49% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[spencer-tb]

Might be good for debugging and failing fast:

Suggested change

	base_cost += gas_costs.NEW_ACCOUNT
	base_cost += gas_costs.NEW_ACCOUNT
	else:
	if metadata["account_new"]:
	raise ValueError("Account new requires value transfer")
	if metadata.get("stipend_returned"):
	raise ValueError(
	"Stipend returned requires value transfer"
	)

[spencer-tb]

Some Amsterdam cases. First one verifies the EIP-8038 repricing. Second one verifies the EIP-8037 dimension split with a new account.

Suggested change

	),
	),
	pytest.param(
	Amsterdam,
	Op.CALL(
	address_warm=False,
	value_transfer=True,
	stipend_returned=True,
	),
	Amsterdam.gas_costs().COLD_ACCOUNT_ACCESS
	+ Amsterdam.gas_costs().CALL_VALUE
	- Amsterdam.gas_costs().CALL_STIPEND,
	id="call_cold_with_value_stipend_returned_amsterdam",
	),
	pytest.param(
	Amsterdam,
	Op.CALL(
	address_warm=False,
	value_transfer=True,
	stipend_returned=True,
	account_new=True,
	),
	Amsterdam.gas_costs().COLD_ACCOUNT_ACCESS
	+ Amsterdam.gas_costs().CALL_VALUE
	- Amsterdam.gas_costs().CALL_STIPEND
	+ Amsterdam.gas_costs().NEW_ACCOUNT,
	id="call_cold_new_account_stipend_returned_amsterdam",
	),

[spencer-tb]

Suggested change

	),
	pytest.param(
	Osaka,
	Op.CALLCODE(
	address_warm=False,
	value_transfer=True,
	stipend_returned=True,
	),
	Osaka.gas_costs().COLD_ACCOUNT_ACCESS
	+ Osaka.gas_costs().CALL_VALUE
	- Osaka.gas_costs().CALL_STIPEND,
	id="callcode_cold_with_value_stipend_returned",
	),

[spencer-tb]

I think we could add a small unit test for returned_stipend_reserve. Small review, I know its in draft 😅

[jochem-brouwer]

(changed target base, need to rebase this one, FYI)