* fixed doc URL for keyspace events

Based on ideas documented in this blog post:

https://www.facebook.com/notes/facebook-engineering/three-optimization-tips-for-c/10151361643253920

The original code was modified to handle signed integers, reformetted to
fit inside the Redis code base, and was stress-tested with a program
in order to validate the implementation against snprintf().

Redis was measured to be measurably faster from the point of view of
clients in real-world operations because of this change, since sometimes
number to string conversion is used extensively (for example every time
a GET results into an integer encoded object to be returned to the
user).

This is just a quickfix, for the nature of the test the right way to fix
it is to average the error of N runs, since otherwise it is always
possible to get a false positive with a bad run, or to minimize too much
this possibility we may end testing with too much "large" error ranges.

This is not a regression but issue #1786 showed the need for this test.

The user @kjmph provided excellent ideas to improve speed of ZUNIONSTORE
(in certain cases by many order of magnitude), together with an
implementation of the ideas.

While the ideas were sounding, the implementation could be improved both
in terms of speed and clearness, so that's my attempt at reimplementing
the speedup proposed, trying to improve by directly using just a
dictionary with an embedded score inside, and reusing the single-pass
aggregate + order-later approach.

Note that you can't apply this commit without applying the previous
commit in this branch that adds a double in the dictEntry value union.

Issue #1786.

In order to make sure every object has its own private LRU counter, when
maxmemory is enabled tryObjectEncoding() does not use the pool of shared
integers. However when the policy is not LRU-based, it does not make
sense to do so, and it is much better to save memory using shared
integers.

PING can now be called with an additional arugment, behaving exactly
like the ECHO command. PING can now also be called in Pub/Sub mode (with
one more more subscriptions to channels / patterns) in order to trigger
the delivery of an asynchronous pong message with the optional payload.

This fixes issue #420.

The code tested many times if a client had active Pub/Sub subscriptions
by checking the length of a list and dictionary where the patterns and
channels are stored. This was substituted with a client flag called
REDIS_PUBSUB that is simpler to test for. Moreover in order to manage
this flag some code was refactored.

This commit is believed to have no effects in the behavior of the
server.

Previously, the command definition for the OBJECT command specified
a minimum of two args (and that it was variadic), which meant that
if you sent this:

OBJECT foo

When cluster was enabled, it would result in an assertion/SEGFAULT
when Redis was attempting to extract keys.

It appears that OBJECT is not variadic, and only ever takes 3 args.

https://gist.github.com/michael-grunder/25960ce1508396d0d36a

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.