Fix race condition bug in NativeMemoryCacheManager

[owenhalpert]

Description

Previously, the window between loading an item from its nativeMemoryContext and acquiring a read lock allowed for the item to be cleaned or evicted before being read for the query. The goal here is to synchronize the load and incRef() operations to avoid this race condition.

We were seeing the following exception when running an OSB vectorsearch workload with 5+ search clients:
[ERROR] search_phase_execution_exception ({'error': {'root_cause': [{'type': 'illegal_state_exception', 'reason': "IndexAllocation-Reference is already closed can't increment refCount current count [0]"}.

With these changes (backported to 2.18 to test on OSB), we are able to consistently execute the entire benchmark successfully with an error rate of 0%.

Thanks to @kotwanikunal for doing the initial investigation and brainstorming on this issue.

Related Issues

Resolves #2262

Check List

• New functionality includes testing.
• New functionality has been documented.
• API changes companion pull request created.
• Commits are signed per the DCO using --signoff.
• Public documentation issue/PR created.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.