Add JDBC persistence backend

[jbonofre]

As discussed during the Polaris Community Meeting, I will add a new plugin implementing JDBC storage backend, without Eclipselink and focusing on PostgreSQL.
It will use "purely" JDBC DataSources with DBCP2 connection pooling.

[jackye1995]

Based on #766 , the JDBC connection needs to support SERIALIZABLE isolation. This would allow a few scalable distributed backend options like CockroachDB and Google Spanner beyond single node SQL engines that could become a bottleneck.

Unfortunately the equivalent product in AWS - Amazon Aurora DSQL, only supports REPEATABLE READ as of today... 😢

[jbonofre]

That's a good point.

If it should work fine for JDBC, I don't think we should limit/focus only on JDBC in Polaris. We have to support also NoSQL.
If we have to change the Polaris data model to support NoSQL, we should discuss about it now, because, it means that we will have to refactore JDBC plugin to support a new data model.

[jackye1995]

We have to support also NoSQL.

Can you explain more about the background on this?

If we want to support both SQL and noSQL, we probably need to create a broad interface that can be implemented by both SQL and noSQL in drastically different ways. I was a bit out of date with where the community is right now, are we acceptable to such change?

[jbonofre]

@jackye1995 I mean that the Polaris data model should allow us to support both SQL and NoSQL backend. The purpose of this issue, the #650 and #844 is to identify the changes needed on the current data model.
So, I propose to work on a doc with:

• problem statement (what are the blockers/issues with the current Polaris data model to support both SQL and NoSQL backends)
• proposals (what the minimal changes needed on the model to support both SQL and NoSQL)

The reason why I think it's important to discuss this today is because it's central to Polaris and any change later will mean to "refactore" the backend plugins.

[jbonofre]

Related to #985

[flyrain]

cc @singhpk234