Fix TestPartitionReader_ShouldNotMissRecordsIfFetchRequestContainPartialFailuresWithConcurrentFetcherIsUsed flakyness

[pracucci]

What this PR does

In #9963 I've introduced TestPartitionReader_ShouldNotMissRecordsIfFetchRequestContainPartialFailuresWithConcurrentFetcherIsUsed. After merging, I realised locally that sometimes the test hang.

I spent 1h debugging it and it's related to the case the kfake control return an error (not a response with error). I'm quite sure the issue has not been introduced with the changes in #9963, but I still haven't understood if the issue is in kfake (can't find a proof there) or an issue in the franz-go library.

I need more time to further investigate it. In the meanwhile I suggest to remove that type of error: the error still make sense due to the error injected by "Simulate a 10% Kafka error rate".

Which issue(s) this PR fixes or relates to

N/A

Checklist

• Tests updated.
• Documentation added.
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
• about-versioning.md updated with experimental features.

[dimitarvdimitrov]

the hanging behaviour looks like the behaviour when returning a nil, nil, true:

mimir/vendor/github.com/twmb/franz-go/pkg/kfake/cluster.go

Lines 467 to 471 in 655c343

	// If the control function returns true, then either the response is written
	// back to the client or, if there the control function returns an error, the
	// client connection is closed. If both returns are nil, then the cluster will
	// loop continuing to read from the client and the client will likely have a
	// read timeout at some point.

[pracucci]

the hanging behaviour looks like the behaviour when returning a nil, nil, true:

Yesteday I investigated it a bit and I'm not 💯 sure about it. In case of nil, err, true kfake should close the connection (without sending any response).