feat(test): add intermediate revert sstore test for 8037

[fselmo]

• Extended the test here to more *CALL* opcodes
• Added intermediate revert test:

---

Claude summary of new test:

The key dimension only the new test covers: the credit is fully deferred in inner (state_gas_used == 0 there, so applied = 0, pending = sstore_state_gas), then bubbles up via incorporate_child_on_success, and lands on a DIFFERENT charge in middle (middle's own slot-1 SSTORE-set, which has nothing to do with
caller's slot-0 set that the refund is keyed to). Middle then REVERTs.

This is the exact code path the PR's state_gas_refund removal rewrites:

• Under the dropped logic: middle.state_gas_refund += sstore when the credit applied in incorporate_child_on_success. Middle's REVERT then triggered caller.state_gas_left += middle.state_gas_used + middle.state_gas_left - middle.state_gas_refund, burning sstore at the error boundary even though caller's matching
  set is the one persisting. The user double-paid.
• Under the new logic: no subtraction. The applied amount stays in middle.state_gas_left and flows to caller via incorporate_child_on_error. Caller's reservoir comes back whole; the matching set in caller stays charged once, as it should.

The other revert tests can't catch this regression: they all have the credit landing in the same frame as its matching charge, so OLD and NEW produce the same answer there. The new test creates the only configuration where the dropped subtraction would have stolen sstore_state_gas from the user's reservoir.

[spencer-tb]

It would be nice to be able to do this within the parameterizer in the future