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.

Now Lua scripts dispatch Redis commands properly calling the call() function. In order to make this possible call() was improved with a new flags argument that controls how the Redis command is executed.

Set a 3.5 GB maxmemory limit with noeviction policy if a 32 bit instance without user-provided memory limits is detected.

Added tests checking ability of the scripting engine to reorder the output of commands with a random output regarding signle elements position in the multi bulk reply.

A few SORT tests made more resistant to false negatives resulitng from poor randomization of Redis hash function with one byte inputs.

Make SORT BY <constant> STORE ... to always produce the same output by force sorting, so that we have deterministic replication of this command.

SORT is now more deterministic: does not accept to compare by score items that have scores not representing a valid double. Also items with the same score are compared lexycographically. At the same time the scripting side introduced the ability to sort the output of SORT when sort uses the BY <constant> optimization, resulting in no specific ordering. Since in this case the user may use GET, and the result of GET can be null, converted into false as Lua data type, this commit also introduces the ability to sort Lua tables containining false, only if the first (faster) attempt at using just table.sort with a single argument fails.

Order output of commands returning random arrays using table.sort when called from Lua, partially fixing issue #165. The issue is yet not completely fixed since we can't add the REDIS_CMD_SORT_FOR_SCRIPT flag in SORT currently, both because it may contain NULLs and because it is not cool to re-sort everything at every call when instead this should be sorted only if BY <constant> is used.

minimal change to obuf-limits.tcl test to make sure there are no false positives with 32bit instances as well.

Lua reports line numbers off by one in error messages

setKey(): call the higher level wrapper setModifiedKey() instead of touchWatchedKey() even if currently they are exactly the same.

SORT with STORE removes key if result is empty. This fixes issue #227.

Fixed a non critical bug signaled by clang static analyzer thanks to Mukund Sivaraman for reporting it: there was a not initialized field populating the cluster message header, but it is always fixed at later time before sending the packet.

patch by Mukund Sivaraman, modified by me to make it simpler and to use
my coding style).

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.