JPA async path is slower now that is synchronous

[michaelabuckley]

Describe the bug
The JPA persistence provider became slower when the async cached query threadpool was removed.

To Reproduce

Steps to reproduce the behaviour:

1. Install 5.3
2. Load some synthea data
3. Make a query that returns more that 500 results. E.g. /Observation?status=active
4. Upgrade to 6.0
5. Compare results
6. The new server is much slower. E.g. 200ms vs 30ms.

Expected behaviour

Sub 50ms response

Environment (please complete the following information):

• HAPI FHIR Version 6.0 vs 5.3
• Postgres

[michaelabuckley]

This one looks related - #866

[jamesagnew]

That #866 suggestion is a really good one.

[michaelabuckley]

This bug got fixed elsewhere by limiting the ES prefetch- #860

See also #536

A modification of #866 is drop the pid, and use a composite PK - (search_pid, search_order). Both of these are known at insert time, and they are small.

Need to test if the inserts are actually batched, and turn that on by default in our props.