feat(rpc): add machine-readable UserOp safety warnings

[realmehmetali]

Summary

• Populate the existing warnings response field with typed, deterministic safety signals.
• Warn on unlimited ERC-20 approvals, active operator approvals, and smart-account authority changes affecting the simulated sender.
• Evaluate the final approval state so a permission revoked or reduced later in the same operation does not leave a stale warning.
• Return no warnings for reverted simulations and deterministically deduplicate warning codes.

Why

Autonomous agents and smart-account wallets need machine-readable safety inputs before they sign. The simulator already extracts objective approval and account-management effects; exposing those facts as stable warning codes lets clients enforce local policy without trusting an external reputation service.

These warnings describe state changes, not intent. They do not label a destination, asset, or operation as malicious.

Safety and scope

• Read-only RPC response change; no consensus, execution, state, or gas behavior changes.
• No external calls, allowlists, reputation feeds, credentials, or new dependencies.
• Only effects on the simulated sender can produce warnings.
• Existing response shape is preserved: the previously reserved warnings array is now populated with typed entries.

Validation

• cargo fmt --all -- --check
• cargo test -p world-chain-rpc --lib (16 passed)
• cargo test -p world-chain-rpc --test fork_simulate --no-run
• cargo clippy -p world-chain-rpc --lib --tests -- -D warnings

New tests cover final approval ordering, grant-then-revoke behavior, account-authority changes, reverted simulations, unrelated accounts, deterministic deduplication, and stable JSON serialization.

---

Note

Low Risk
Read-only RPC response enrichment with no execution or state changes; only affects clients consuming the previously empty warnings array.

Overview
The simulate_unsignedUserOp response fills the reserved warnings field with typed SimulationWarning entries instead of always returning an empty array. Warnings describe objective effects on the simulated sender only—unlimited ERC-20 approvals, active ApprovalForAll-style operator grants, and smart-account authority changes (ownership, modules, proxy upgrade, self-destruct, authorization)—not malicious intent.

generate_simulation_warnings derives warnings from existing exposure_changes and contract_management output. Approval signals use the final state per (asset, counterparty, approval kind) so a later reduce/revoke in the same UserOp does not leave a stale warning. Reverted simulations return no warnings; results are deduplicated and ordered via BTreeSet.

New unit tests cover approval ordering, grant/revoke, account-management mapping, unrelated accounts, reverts, and stable JSON serialization for warning types.

^{Reviewed by Cursor Bugbot for commit ee6bb2a. Bugbot is set up for automated code reviews on this repo. Configure here.}