Adding ROLE command

[prvyk]

An initial attempt to add the ROLE command + docs + tests. Likely this will need some revisions.

https://redis.io/docs/latest/commands/role/

1. After Fix append after expiration #961 the new ACL test blows up (it passed previously). I suspect that introduced an unrelated regression affecting this test, since we just use normal serialization methods.
2. I'm not sure the replication state check is correct at all. Note we check the new IsPrimary() since the recovering case should go to the replica branch unlike normal IsReplica().
3. Didn't mention 'sentinel' possibility in docs, as Garnet does not support this mode.
4. Far from sure this is the best code organization, it seemed cleaner than doing the logic in one method.
5. Minor effect on Metrics code path (see first commit), but with same result.

[badrishc]

I don't think the unit test error is related to #961. It might be related to #925. The issue is that GarnetClient cannot handle an array response where one of the array elements is itself another array. It is possible that prior to #925, we were swallowing this somehow?

RECV: [*1|$4|ROLE|]
SEND: [*3|$6|master|:0|*0|]
Exception thrown: 'Garnet.common.Parsing.RespParsingException' in Garnet.common.dll
Unexpected character '*'.

One solution would be to add basic support for an array field in RespReadResponseUtils.cs line 192, something like this:

                else if (*ptr == ':')
                {
                    if (!TryReadIntegerAsString(out result[i], ref ptr, end))
                        return false;
                }
                else if (*ptr == '*')
                {
                    if (!TryReadStringArrayWithLengthHeader(out var subArray, ref ptr, end))
                        return false;
                    result[i] = string.Join(", ", subArray);
                }
                else
                {
                    RespParsingException.ThrowUnexpectedToken(*ptr);
                }

[prvyk]

Thanks. Added this to unblock review for now. Perhaps this should be fixed separately, if it is it would be easy to revert/force push/ to get that out from the PR.

[badrishc]

Feel free to split this change out to its own PR. It is not perfect, but we don't really plan to make GarnetClient perfect for all cases anyway. it just needs to handle these nestings gracefully.

[vazois]

Great work! Some minor comments

[prvyk]

The new CLUSTER NODES test has an interesting Linux-only failure... CLUSTER NODES is unrelated to this PR, I think it should get its own?

[vazois]

The new CLUSTER NODES test has an interesting Linux-only failure... CLUSTER NODES is unrelated to this PR, I think it should get its own?

Yes, revert the change it is not related to this PR. If you would like to address the issue in a separate PR please go ahead.

[vazois]

I just noticed some more things. Please address.