Previous design was using extra threads needlessly and had some
hacky data sharing going on. Slim this down to just two classes
and utilize threads appropriately.

There's no need to use a GLConsumer unless we really want to draw
on the surface or send it to a hardware encoder. Get rid of all
this stuff and just use CpuConsumer.

Put the VNCFlinger in the screenData so that it can be accessed
by clients. This removes the need for the message queue. The
buffer ready event was removed in a previous commit.