Better pool connection fail handling

As per @kr suggestion, if a connection in the pool has an error, it's probably best to close
it, then attempt to reconnect periodically.

Also perhaps be able to add new connections to the pool as well?

No file, test, or entry point is named. Start by locating the pool's connection error handling and existing connection-management tests, then clarify whether failed connections should be closed and retried and whether the pool may grow. Done means the agreed behavior is implemented and covered by tests.

Written by the indexing model from the issue text.