Fixes to c->reply_bytes computation, and debug messages to closely study the behavior of memory pressure + slaves + maxmemory + blocked slaves.

Precision of getClientOutputBufferMemoryUsage() greatily improved, see issue #327 for more information.

1) sendReplyToClient() now no longer stops transferring data to a single
client in the case we are out of memory (maxmemory-wise).

2) in processCommand() the idea of we being out of memory is no longer
the naive zmalloc_used_memory() > server.maxmemory. To say if we can
accept or not write queries is up to the return value of
freeMemoryIfNeeded(), that has full control about that.

3) freeMemoryIfNeeded() now does its math without considering output
buffers size. But at the same time it can't let the output buffers to
put us too much outside the max memory limit, so at the same time it
makes sure there is enough effort into delivering the output buffers to
the slaves, calling the write handler directly.

This three changes are the result of many tests, I found (partially
empirically) that is the best way to address the problem, but maybe
we'll find better solutions in the future.

Use less memory when emitting the protocol, by using more shared objects for commonly emitted parts of the protocol.

Client output buffer limits: configuration of parameters for the different classes of clients implemented.

asyncCloseClientOnOutputBufferLimitReached() now ignores clients with REDIS_CLOSE_ASAP flag already set. Return value of the function changed from int to void since it is not used. Fixed logging of the client scheduled to be closed.

Implementation of the internals that make possible to terminate clients overcoming configured output buffer (soft and hard) limits.

Track the length of the client pending output buffers (still to transfer) in a new field in the client structure.

On crash print information about the current client (if any), command vector, and object associated to first argument assuming it is a key.

Protections against protocol desyncs, leading to infinite query buffer growing, due to nul-terms in specific bytes of the request or indefinitely long multi bulk or bulk count strings without newlines. This bug is related to Issue #141 as well.

Fixed replication when multiple slaves are attaching at the same time. The output buffer was not copied correctly between slaves. This fixes issue #141.

Fixed memleak in CLIENT INFO, added simple test that will work as regression test on mac os x and in the CI when running over valgrind. This fixes issue #256

minor refactoring to networking.c adding a separated function to get a string representing the current state of all the connected clients.

new counter in INFO output: rejected_connections with number of dropped connections because of maxclients limit reached.

code generating the CLIENT LIST output refactored to have a function that is able to render a single client into a client info string.

Multi bulk optimization for creating big objects without copying data is no longer optional, #ifdefs removed. Also debugging messages removed.

further optimizations for the multi bulk protocol parsing code when big objects are transmitted to Redis.

sdsIncrLen() / sdsMakeRoomFor() used to avoid copying to intermediate buffer while reading the client query.

maxclients configuration is now implemented dealing with the actual process rlimits. Setting maxclients to 0 no longer makes sense and is now invalid, the new default is 10000.
See issue #162 for more information.