fix wheel impulse scaling

[tomo0613]

#946

explanation by Claude:

The Bug: A misplaced gate in updateFriction

In update_friction(), after determining that sliding == true (the combined forward+side impulse exceeds the friction circle), there's a correction loop that scales both impulses down by skid_info. ...

// before
if wheel.side_impulse != 0.0 && wheel.skid_info < 1.0 {

// after
if wheel.skid_info < 1.0 {

The wheel.side_impulse != 0.0 guard is the whole problem. It was presumably written as a micro-optimisation ("skip wheels with no lateral load"), but it incorrectly prevents forward_impulse from being scaled down when braking straight — since side_impulse is exactly 0.0 in that case, the entire body of the if is skipped, and the braking impulse is applied without any friction-circle capping.
The skid_info < 1.0 check alone is the correct and sufficient guard — it's only < 1.0 when the friction circle is actually exceeded, so there's no wasted work. Multiplying side_impulse (which is already 0.0) by skid_info is a no-op, so removing the outer guard has zero downside.

This bug exists in the original Bullet source (lines 675–682 of btRaycastVehicle.cpp) and was presumably never caught because most demos and tutorials involve some steering before heavy braking — the exact scenario that masks it.