Currently the recording will include all queries, including connection setup "hello" etc.

The replay client will generate some of these itself, and the behaviour of these once-per-connection queries is not particularly relevant to the workload which follows.

On the persistence thread, incoming queries handed off from the query-servicing threads should be filtered to only queries we wish to replay.

This is likely easiest to implement as an allowlist - enumerating interesting query types like "find", "findAndModify" etc. is likely easier than the inverse blocklist.

Note that this does not need to filter to exactly the queries we are currently able to replay; recordings can and should continue to include all queries we intend to replay, even if we do not yet support them. This means replaying a particular workload can call out any queries which are seen but cannot be replayed