feat: Enable users to use Entity Query as CTE during historical retrieval

[blaketastic2]

What this PR does / why we need it:

This PR gives users the option to use CTEs for querying historical features instead of the current approach of creating a table, using the table and then dropping the table.

The option sits behind a flag that's configurable on the Offline Store configuration.

Which issue(s) this PR fixes:

When trying to join more than 1 or 2 entities to a feature view, we started noticing performance issues. During this time, we were able to manipulate the query SQL to use a CTE instead of the TABLE created by default, and saw noticeable improvements in getting results back.

We believe this will have limited if any downsides to the existing implementation.

Misc

There's a note that this SQL template should stay in sync w/ Redshift, but we are now drifting.

[franciscojavierarceo]

Awesome, would you mind updating the documentation for Postgres while you're here?

[blaketastic2]

@franciscojavierarceo Let me know if the updated docs are what you were looking for.

[franciscojavierarceo]

this lgtm, as a follow up you may want to have a way to enforce the temp_table to be unique if two people end up having collisions with the desire to operate on different partitions of the data.

[blaketastic2]

I believe in this particular case, the entity_query CTE will run without the need to worry about partitioning. Also, it looks like a flaky test failure again.