Netty 4.1.87 to 4.1.94 regressed memory usage

#10401 is fixed, but we're [seeing ~1.8x increased memory usage](https://grafana-dot-grpc-testing.appspot.com/d/Qb8s8b6nz/psm-dataplane-performance-on-gke-atupstream-master?orgId=1&from=1686810052419&to=1689961252419) by a benchmark that compares proxyless with Envoy (look at Memory_java).

The only interesting thing when it increased was the Netty upgrade. However, it didn't recover when we downgraded to 4.1.93.Final.

We don't yet have a theory on how proxyless was impacted but not grpc+Envoy.

Start with the linked Grafana benchmark and compare the Memory_java results for proxyless and Envoy. Review the Netty changes between 4.1.87 and 4.1.94, then verify whether 4.1.93.Final reproduces the increase. Done means identifying the regression cause and confirming memory usage returns to the expected range without affecting the Envoy comparison.

Scritto dal modello di indicizzazione a partire dal testo della issue.