gRPC access log has no back pressure in case if backend can not keep up with the load. In this case it starts buffering all entries without any upper bound.

In a sense it is similar to #9458 but for other types of loggers.

In order to avoid OOM situations Envoy should:

1. Limit memory usage of buffered entries.
2. Drop new entries after some threshold (which can default to smth like 1MB).
3. Emit metric for number of dropped log entries.