fix: expand memory before slicing initcode in CREATE2 wrapper

[RealiCZ]

Forward-port of #293 onto main. Same diff, same intent — bringing the CREATE2 memory-expansion fix into the active development line. #293 was originally merged onto release-v1.5.2 (based on v1.5.1), this PR carries it forward without the 1.5.2 bump or other release-branch commits.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.5%. Comparing base (09ceb5d) to head (af2b1c8).

☔ View full report in Codecov by Sentry.
📢 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.

[claude]

Review: LGTM

The fix is correct and well-tested. A few notes from the review:

Correctness

• resize_memory! on the OOG path returns early from the function (via the macro's internal return), so slice_len is never reached on unallocated memory — that is exactly the bug being fixed.
• Deferred compute_gas! recording is the right approach: the canonical compute_gas_ext::create2 path captures gas via a gas_before/gas_after window, but when this wrapper's resize_memory! runs first, the inner revm resize_memory! becomes a no-op, causing that window to silently miss the expansion gas. Recording it separately after a successful inner CREATE2 fills that gap correctly.
• All three leakage paths from CLAUDE.md are handled: (1) inner CREATE2 OOGs → run_inner_instruction_or_abort! skips the deferred recording, (2) compute gas limit exceeded by resize_gas → interpreter halts after the fact, consistent with other compute gas halts, (3) resize_gas == 0 short-circuits the borrow entirely.

Tests

The three tests cover the three important scenarios cleanly: no panic, correct accounting on success, and correct skip on OOG. The < 50 delta tolerances are generous relative to the actual differences (~1-3 gas for a PUSH opcode), but that is fine.

No issues to block merge.