[prometheusremotewriteexporter] reduce allocations when serializing protobufs

[edma2]

Description:

There are two allocations that happen on every push

1. Serializing the remote write protobuf to a byte array.
2. Compressing the byte array with snappy.

Since these buffers can be quite large, we can save some allocations with a sync.Pool.

Link to tracking Issue:

Testing:

Tests still pass. We have been running this successfully in production for a few months now.

Documentation:

[dashpole]

Looks great! Are you able to add a benchmark to demonstrate the improvement and make sure it doesn't regress in the future?

cc @jmichalek132 @ArthurSens feel free to review as well

[jmichalek132]

LGTM, but could you please look at the failing compliance tests? So turns out that should be fixed now #35119 so rebasing should hopefully make them pass.

By any chance is there any visible improvement you can see with this change from the metrics of the collector, if so could you post them here?