Improve reliability of transaction inclusion/resume functionality

The state does not get updated with the hashes of in-flight transactions. This can lead to frustrating errors such as:

- Old transactions sticking around in the mempool and preventing future transactions from being added.
- Transactions timing out in the first run of `op-deployer`, but succeeding afterwards. This leads to the deployment having to needlessly start from the beginning.

To resolve, we need to:

1. Write intermediate transaction hashes to state once they are broadcasted.
2. On subsequent runs, check if the transaction hashes are in the mempool or confirmed so that the deployment can continue.

Start by locating the op-deployer transaction broadcast, state persistence, and resume paths in the Go code. Trace how intermediate hashes are recorded and how later runs inspect mempool or confirmed transactions. Done means deployments can resume from recorded in-flight transactions without retaining obsolete mempool entries or restarting unnecessarily.

Written by the indexing model from the issue text.