The agent callback is never called when making a 'dcp-vbtakeover' stat call to a multi-node cluster

When making a dcp-vbtakeover stat call to a multi-node cluster we expect that some of the responses to be NotMyVBucket, however, gocbcore absorbs this error before successfully calling the agents callback function. In the case of backup, this is leading to timeout errors as seen in MB-35905.

Steps to reproduce:
1) Spin up a cluster of 4 instances of Couchbase Server.
2) Provision three data nodes and one node running all other services.
3) Load one of the sample-buckets.
4) Run the provided script stats-call-test.go.
5) You should see that the callback function is never called.

Expected results:
The stats call should succeed and the agents callback function should be run.

Actual results:
The callback is never run leaving the agent to handle timeouts.