fix: route the impersonated fork signer through the nonce queue

[apexearth]

getSigner() wraps the KMS and private-key branches in maybeWrap(), but branch 3 does not:

Branch	Wrapped
1. AWS KMS	return maybeWrap(new DirectKmsTransactionSigner(...)) ✅
2. DEPLOYER_PK / GOVERNOR_PK	return maybeWrap(new ethers.Wallet(pk, provider)) ✅
3. FORK=true + IMPERSONATE	return provider.getSigner(address) ❌

Without the wrapper, sendTransaction never routes through wrapSignerWithNonceQueueV5, so Talos does not record the transaction. The visible symptom is a run that completes successfully with an empty transaction list.

-    return provider.getSigner(address);
+    return maybeWrap(provider.getSigner(address));

No await — getSigner() is synchronous on ethers v5 (5.7.2 here), unlike v6.

DATABASE_URL is present in the runner container, so maybeWrap() does attach the queue — nothing else was missing.

Why now

oplabs/talos#30 makes an anvil mainnet fork with an impersonated relayer the default dev environment, so branch 3 goes from rarely-exercised to the everyday path. Companion fix in arm-oeth: OriginProtocol/arm-oeth#339.

Worth a reviewer's judgement

The nonce queue does real work — rebroadcasts, fee bumps, replacements, driven by the NONCE_QUEUE_* settings. On a fork that is arguably the point (you exercise the production path), but dev transactions become managed rather than fire-and-forget. If you would rather impersonated runs stay lightweight, the alternative is recording them without the queue.

Not verified

I could not typecheck — the worktree has no node_modules. The change is a one-line call into maybeWrap(), already used twice in the same function, so it should be safe, but CI is the check.

[apexearth]

merge if you approve