The cg3d.h and cg3d.C files were my attempt at creating a
set of generic shape-intersection routines.  They've
largely failed, due to the numerical difficulties outlined
below; we now use the Magic Software routines in the "mgc" 
directory.

------------------- Numerical Difficulties: ------------------

src->src works fine (error identically 0)

make OPTS=-g test_transfer && gdb ./test_transfer
r test_transfer_dir/src.noboite test_transfer_dir/dest.noboite

Problem:
 True volume 3.124847, but total is only 3.115913 (err -0.008934)
 tet 284: 82 97 9 122
 (-0.052, -0.052, -4.500) (-0.052, 2.465, -5.017) (-0.052, -0.052, -7.500) (-2.535, -0.052, -5.017)

-----------
Reason: Blatant non-manifold shape:
s=9, srcMesh=@0xbffff970, d=246, destMesh=@0xbffff950, 
    dest=0xbffff1a0
Corner volume: 1.5844484919394499
(-1,0,0) volume: -2.8493726524263607

Source Tetrahedral mesh: 15 tets, 11 points
Destination Tetrahedral mesh: 479 tets, 130 points
Volume3dDest: swept volume is not manifold (1.584448 vs -2.849373)
dest tet 246:  tet 246: 127 35 128 76
 (-0.052, -0.052, 4.500) (-0.052, -0.052, 7.500) (2.465, -0.052, 4.983) (-0.052, 2.465, 4.983)
src tet 9:  tet 9: 2 10 9 4
 (7.500, 7.500, -7.500) (7.500, -0.052, -0.052) (-0.052, -0.052, 7.500) (-0.052, -0.052, -7.500)

 -> This is *not* a symmetry-breaking problem (swapping src and dest has no effect).

It does appears to be roundoff-related, since it appears for epsilon e-12 through e-8,
but not e-7 through e-4.

Running with e-8 epsilon causes one huge non-manifold shape, but it doesn't appear 
to matter in the end.
Running with e-7 epsilon causes no errors.
Running with e-6 epsilon causes numerous tiny non-manifold shapes and volume errors.

  -> Roundoff is a significant problem.

Changing the halfspace definition from (n dot p) + d for a scalar d
to n dot (p-o) for a vector o significantly improves roundoff:
with the friendly halfspace, epsilon can be made as small as e-11 without
any problems.  e-12 still causes big nonmanifold errors.

With OPTS="-O", we still get problems, even at e-8 (e-7 happens to work, tho').