Throughput drops to 0 on Subdoc workload on rebalancing cluster

Running the SDKD jobs, a number of the subdoc workload tests have begun to fail:

http://sdkqe-testresults.couchbase.com.s3.amazonaws.com/SDK-SDK/CB-7.2.4-7070/Rb2OutEpt-SUBDOC/03-08-24/050401/d6247cae2c76fdaa338a1ac449e833ef-SD.html

.NET Logs:

SdkdConsole.log

In this run, the cluster starts with nodes: http://172.23.109.150:8091, http://172.23.109.146:8091, http://172.23.109.144:8091, http://172.23.109.142:8091 

And these 2 are rebalanced out:  http://172.23.109.150:8091, http://172.23.109.146:8091

Note any rebalance in or out of the cluster has the same effect.

Looks like lots of `Couchbase.Core.Exceptions.KeyValue.DocumentNotFoundException` errors. This is see with 7.2 and 7.1, am checking 7.6.

Would also add that in the KV tests, the performance has also gotten worse with failing operations where previously we didnt have any failures:

RC2 : http://sdkqe-testresults.couchbase.com.s3.amazonaws.com/SDK-SDK/CB-7.2.4-7070/Rb2Out-KV/03-08-24/036385/44fd280c00f44874e943f63da3dc8a34-MC.html

55e03c59: http://sdkqe-testresults.couchbase.com.s3.amazonaws.com/SDK-SDK/CB-7.2.4-7070/Rb2Out-KV/03-07-24/073139/1df353d7854c6efbbdf3fcf2835cf4cc-MC.html

Though in these tests, the operations do seem to recover after some time, where they do not in the subdoc tests.

Note I had run some tests last night in prep, and it was passing with .NET commit 
55e03c598bce6bc7fd5814856b377720a87206e6
So it was something in the last 3 commits since then.

Passing test logs from 55e03c598bce: (if useful)
SdkdConsole-passing.log.zip

Edit: Matthew Bray Has also seen DocumentNotFound errors in FIT-SIT Capella tests - i.e. KV ops against a capella cluster that undergoes any scale.

https://performance-sdk.couchbase.com:8080/situationalSingle?situationalRunId=f8c941ac-216d-4843-85f9-8ba1615a06f0&runId=958d34d5-bfc3-4a6b-9625-1f86acec8ff0

So if wanting to reproduce this might be the easiest way.