cl_movement 0 should no longer be causing an input replay (which was slow and pointle...

cl_movement 0 should no longer be causing an input replay (which was slow and pointless), though I don't know how it was doing so before...
added code to prevent pogostick jumping in the cl_movement prediction (since the qc prevents pogostick jumping, it was inaccurate to simulate it clientside)
changed view/gun bobbing to use cl.movement_velocity for the xyspeed math, this greatly improves the clientside prediction

git-svn-id: svn://svn.icculus.org/twilight/trunk/darkplaces@6144 d7cf8633-e32d-0410-b094-e92efae38249