Previously we returned an error.

This way as a side effect of running the test we also stress the latency
monitor data collection.

This happens in the child process so we don't care about latency:
better to sync metadata as well.

This commit adds both support for redis.conf and CONFIG SET/GET.

It is not a good idea to bloat the code with gettimeofday() calls if the
instance is working well, and turning monitoring on at runtime is a
joke.

Instead we update the old sample with the new latency if it is greater.

We introduce the distinction between slow and fast commands since those
are two different sources of latency. An O(1) or O(log N) command without
side effects (can't trigger deletion of large objects as a side effect of
its execution) if delayed is a symptom of inherent latency of the system.

A non-fast command (commands that may run large O(N) computations) if
delayed may just mean that the user is executing slow operations.

The advices LATENCY should provide in this two different cases are
different, so we log the two classes of commands in a separated way.

Introduced in Redis 2.8.10 because of a change in Sentinel.
This closes issue #1837.