Teach the `PendingPayableScanner` on pending payble management

[bertllll]

fill me in later

[bertllll]

The PendingTooLong issue

Use caching of those transactions that end with this failure result so that you can have a structured list of the txs that you are obliged to doublecheck.

The steps should be: a) rewrite the newer (successor) record in sent_payable with the nonce N by the data of the older transaction, now read from the failed_payable table. I say rewrite because I want to avoid the db state inconsistency precisely, That means, I'd like to have a single method to call, which will internally perform: the old row deletion + insertion of the new row, but committed no earlier than both of them done. I say we should use an explicit rusqlite tx, that gives you the control over when it is committed. However, you may be able to work it around by writing an UPDATE sql which will update the whole row though (every valuei in the row.) I wonder if that is considered a hack in the community of the data-engineers? (According to the ChatGPT, it isn't a nonsense. I'd then highly recommend that as it is much simpler to implement) b) delete the record from failed_payble