replaced cl.movement_queue with direct use of cl.movecmd array, this

replaced cl.movement_queue with direct use of cl.movecmd array, this
saves a little memory but mostly it simplifies some logic
reworked much of the prediction system
now extrapolates latest movement gathered instead of interpolating
between the last two moves sent, this reduces visual latency by 50-99ms
and deals better (but not perfectly) with mismatched framerates like
cl_maxfps 30 and cl_netfps 20

git-svn-id: svn://svn.icculus.org/twilight/trunk/darkplaces@8208 d7cf8633-e32d-0410-b094-e92efae38249