refactor(eip8037): defer state-gas to frame return via NewStateTracker

[rakita]

Summary

Switches EIP-8037 state-gas from immediate charging to a deferred model. State-creating opcodes bump a simple per-frame counter on the Interpreter (new_state); the totals are reconciled with the gas tracker at frame return inside EthFrame::process_next_action.

• on ok: state_gas_refunded added to the reservoir first, then state_gas charged (may OOG → result flipped to OutOfGas before journal commit)
• on revert/halt: counters dropped — state work didn't happen, no reservoir change

Notes

• NewStateTracker (crates/context/interface/src/cfg/new_state_tracker.rs) is intentionally minimal: two u64 counters and a small set of add_* / remove_* methods. Uses hardcoded Glamsterdam EIP-8037 values (bytes_per_unit × CPSB_GLAMSTERDAM) for now — does not read GasParams at runtime.
• Migrates the immediate-charge call sites (SSTORE in host.rs, SELFDESTRUCT in host.rs, CREATE/CREATE2 in contract.rs, CALL-with-value in call_helpers.rs) to bump interpreter.new_state.add_*(). Removes the state_gas! macro.
• return_create no longer commits/reverts the journal; it validates code, charges regular code-deposit + hash gas, bumps add_code_deposit_bytes, and lets process_next_action apply state-gas and finalize (set_code on success).
• handle_reservoir_remaining_gas is simplified — the deferred model removes the spill-on-revert distortion, so the old logical_refund / baseline / excess math is no longer needed.
• Parent's return_result Create-branch: on child failure → interpreter.new_state.remove_create_account() (replaces refill_reservoir(create_state_gas)).

Test plan

• cargo build --workspace clean
• 23 EIP-8037 tests in revm-ee-tests currently fail because the test suite overrides sstore_set_state_gas etc. via GasParams (e.g. 200_000) but the new NewStateTracker ignores those — the numbers it produces are EIP-8037 Glamsterdam defaults (e.g. SSTORE = 32 × 1174 = 37_568). Need to either update test expectations, switch NewStateTracker to read GasParams/cpsb, or both.
• Spot-check tracing/inspector output (state_gas no longer goes negative since charges are deferred)

[codspeed-hq]

Merging this PR will degrade performance by 4.84%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 8 regressed benchmarks
✅ 168 untouched benchmarks
⏩ 1 skipped benchmark¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	REVERT_50	109 µs	113.1 µs	-3.59%
❌	Simulation	SLOAD_50	24.3 µs	25.4 µs	-4.3%
❌	Simulation	SSTORE_50	30.2 µs	31.5 µs	-4.08%
❌	Simulation	subcall_1000_transfer_1wei	1.2 ms	1.3 ms	-4.84%
❌	Simulation	subcall_1000_same_account	1.1 ms	1.2 ms	-3.3%
❌	Simulation	transact_1000tx_commit_inner_every_40	2.2 ms	2.4 ms	-4.37%
❌	Simulation	subcall_1000_nested	2.2 ms	2.2 ms	-3.43%
❌	Simulation	transact_commit_1000txs	2.8 ms	2.9 ms	-3.97%

---

_{Comparing dev5-new-way (05c0d7a) with main (bd89862)²}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

2. No successful run was found on snobal-devnet-4 (d9b5182) during the generation of this report, so main (bd89862) was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩