fix: wrap spoke pool indexer in a sql transaction

[amateima]

This PR wraps the execution of SpokePoolIndexerDataHandler.processBlockRange() in a SQL transaction. In case of throwing an error, the database should be reverted to the state prior to the block range being processed.

• fetchEventsByRange is not included in the SQL transaction
• all queries must be executed using the new transactionalEntityManager
• to ensure backwards compatibility, in some functions transactionalEntityManager is marked as optional. In such cases the queries are executed using the DataSource instance
  Example:

const repository = transactionalEntityManager
      ? transactionalEntityManager.getRepository(entity)
      : this.postgres.getRepository(entity);

• managed advisory locks pg_advisory_xact_lock that gets released automatically after the transaction ends get replaced with session lived pg_advisory_lock which needs to be unlocked explicitly
  The following pattern has been using in the SpokePoolProcessor:

try {
     // Acquire a lock to prevent concurrent modifications on the same relayHash
     call pg_advisory_lock(),
} finally {
    // Explicitly call unlock for both successful or failed execution
     call pg_advisory_unlock()
}

[linear]

ACX-4037 Wrap `SpokePoolIndexerDataHandler` in a SQL transaction

[james-a-morris]

Any chance we can make a singleton for this EntityManager? Without it we need to pass this variable down to a lot of inner functions

[amateima]

I don't think it's possible to make a singleton because this instance must be limited to the scope of the callback passed as a parameter to createTransaction() function

[daywiss]

great work!

[daywiss]

this makes sense to make it compatible with tests, but i wonder if we should somehow require this when running for real, i dont think we ever want this to run without it in prod

[amateima]

I agree, I need to find a way to make it required. It's easy for someone less familiar with the code to miss using it