small bugfix to ShapeBase::getCameraTransform()

this is present in TGE 1.4.2, and probably in TGE 1.5 and TGEA and will probably be present in T3D as well.

ShapeBase::getCameraTransform() is responsible for determining where the camera should be positioned given a shapeBase. ie, the typical over-the-shoulder view.

this task includes making sure that the camera doesn't get backed-up into a wall or something. to do that, the routine casts a ray from the shapeBase object to the target eyepoint, and if it hits something, it moves the eyepoint to "just in front of" the the collision point.

this "just in front of" amount is calculated as follows (code has been simplified)

// project vec onto the normal: (divide by veclen because it may be != 1)
adjust = 0.1 * dot(vec, normal) / veclen;

and used later to move the camera away from the collision like this:

eyePointFinal = eyePointTarget - (vec * adjust);

see the problem ?
if vec is not unit-lengthed, then vec * adjust is going to vary with the length of vec, when what we really want is an offset from the wall which doesn't vary with the length of the target eyeVector.

how does this affect you ?
probably not at all,
since in stock TGE the camera distance is always either 0 or 1.
but if you should want to let the user adjust the camera distance (say for filming machinima), and the camera is backed up against a wall, the final eyePoint will vary as the target camera distance changes. which looks funny.

the solution is to divide the adjust term one more time by vecLen, so that their product is independant of vec's length.

along the way we get to eliminate a square root as well. always satisfying.

shapeBase.cc
change these two lines:

F32 veclen = vec.len();
         F32 adj = (-mDot(vec, collision.normal) / veclen) * 0.1;

to this:

F32 vecLenSq = vec.lenSquared();
         F32 adj = (-mDot(vec, collision.normal) / vecLenSq) * 0.1f;