RDB Loader: manage several parallel connections

[chuwy]

For #457 we need to maintain at least two connections in parallel: one for primary data loading workflow, another for periodical checks.

We considered common solutions like Hikari CP, but they're tailored for specific OLTP use cases and don't provide (or at least I couldn't find) complex connection acquisition logic, like backoff period (when cluster is on maintainance) or prior SSH tunneling.