QXcbBackingStore performance improvements with regards to SHM

Re-introducing SHM pixmaps should net a performance improvement on Intel hardware:

13:04 < sroedal> is ShmPixmap vs ShmImage going to be a big gain?
13:07 < ickle> yes, should net another factor of 5 in reduction of cpu overhead
13:08 < ickle> (given kernel support)
13:09 < ickle> effectively it gives you zero copy into the gpu, rather than just zero copy into the xserver address space
13:10 < ickle> (depending on various factors, ofc)

Also, instead of doing a sync() we could track the request / reply sequence numbers to check if a put image request has already been processed.