mouse samples.  (Actually don't believe delta-t smaller than .15 sec.)

and let postdraw return an int.  Keep looping, re-rendering,
until postdraw returns zero.

have "o2w" transforms that are really interpreted relative to camera space.
Also start adding per-object clip support; this requires change to
subcameras, supplying projection parameters subc_lrbt()
rather than a (clip-dependent) projection matrix.

for good accuracy throughout the range.

Tidy up messages -- don't append \n in msg() calls.