Fix FindCoordinatorResponse.encode to allow nil Coordinator

[horkhe]

FindCoordinatorResponse.encode function panics if Coordinator is nil, so I fixed that. Besides I added MockFindCoordinatorResponse to use in tests.

[eapache]

This is not matched by the fact that we hard-code passing version 0 into Coordinator.encode?

[buyology]

Hm. We are currently sharing the same decode/encode method for Metadata* and FindCoordinator* but since version 1 of the MetadataResponse they are not the equal.

Metadata has rack whereas FindCoordinator does not — so actually 69-73 can be dropped

[eapache]

It would be nice if you could add a test for this case.

[buyology]

Good catch. Maybe we should line up the test and encoding of a nil coordinator with the actual response where NO_NODE is defined like so:

    private static final Node NO_NODE = new Node(-1, "", -1);

[horkhe]

@eapache @buyology done. Version is not properly used in encoding/decoding. NoNode was introduced and used in encoding if Coordinator is nil. The test covers both versions.

[buyology]

Sorry if I was unclear about my comment re: the shared decode/encode.

The short of it was that the FindCoordinatorRequest/Response does not have a rack-field:
https://kafka.apache.org/protocol#The_Messages_FindCoordinator

Sending a followup patch so let's continue there

[horkhe]

Good catch, thank you @buyology.