YaGL: Transport improved
The improvements are:
* No more mlock/munlock. We now have ioctls for locking drawable
memory so that compile transfers could be created on host. This
is only used by offscreen backend though
* We're now using a single buffer for marshalling instead of two: one
for commands and one for data. Also, the buffer can now be of any
size up to 2M, it's implemented as a page list, thus, we don't need
to allocate contigous memory anymore
Change-Id: Ia9b716c9135df75535dc515367550c9fbcf9c737