stats: use a release-assert when histogram-merging tries to run concurrently

[jmarantz]

Description:

Converts an ASSERT to a RELEASE_ASSERT, as it could be related to memory corruption observed in #3223 . This was originally an if-statement blocking the corruption in @ramaraochavali histogram PR but per @mrice32 we changed it to an ASSERT, which means if it happens in production, it's a silent corruption.

Risk Level: Low

Testing: //test/... pass. but test/extensions/stats_sinks/metrics_service:metrics_service_integration_test fails with this assertion when run under valgrind.

Docs Changes: N/A

Release Notes: N/A

[mattklein123]

Can we hold on merging this for a little bit. In general, we don't have asserts like this, as it's a program invariant that shouldn't happen and we should fix the underlying issue. I already have some ideas on what might be wrong but I'm still poking.

[jmarantz]

i'm in no rush :) Definitely agree we should fix the underlying issue that allows this to occur.

[mattklein123]

@jmarantz I'm going to close this for now. Let's reopen later if we think we really need this. Would rather just keep debugging for now.