fix: disable vt proto pools

[kolesnikovae]

Should fix #3024

I tested distributors without VT protobuf pools and I think that it's probably better to disable them altogether:

• We do not benefit from reduced allocs as it's compensated by the cost of sync pools and resets (CPU time).
• Memory consumption is 10-15% lower without pools.
• Code is simpler. I suspect that Excessive memory consumption in distributors causes OOM #3024 is caused by wrong handling of pooled items. By removing the pools we eliminate this factor.

Version w/o pools deployed at ~10:10:

The impact on GC is understood and expected:

[aleks-p]

The code is simpler and the impact looks good. I only left one remark about a change in a generated file.

[simonswine]

Thanks for getting some real numbers on this ❤️

LGTM, after we adapt the generation

[kolesnikovae]

Yeah, thanks for pointing this out guys! That's a leftover – we had a dependency in the JFR parser, so I hacked it quickly for testing and forgot to rollback. I removed the dependency and the generated code for pools. Next step is to write/revive our very own efficient pprof parser