Improve database reconnect loop behavior when connectivity configuration is incorrect

[amotl]

Problem

At crate/cratedb-examples#174 (comment) ff, we discussed a problem that MLflow, when configured with an invalid database connection string, would still go into the reconnect loop, hogging it forever, because, well, it can't recover from a configuration error without further ado.

Suggestion

Improve the situation by either improving the reconnect loop itself, and/or by probing the database connection before actually starting the MLflow service/subsystem. If that fails, the daemon program should terminate.

Let's see how that fits into the other use case, when MLflow is embedded into the application itself, but it should work likewise, mainly preventing to go into the reconnect loop forever when it does not make any sense at all.