Log appender sharding key should be used as Kafka record key

Today org.nuxeo.lib.stream.log.LogAppender#append(java.lang.String, M) is appending a record to a Log using a hash of the provided key, the key itself is not saved.

When using Kafka we could persist the key as the Kafka key record.

This is not the case today because the notion of tuple (key, record) does not exists in Chronicle Queue, therefore if we need a key we encode it inside the record itself.

The interest of having a Kafka key is when we want to play with log compaction or interact with other systems that rely on the Kafka tuple (key, record).

This is also make sense for NXP-24512 so we can compact the latencies topic.