Ignore Bullet collision contact points with distance = 0

[madmiraal]

Currently, Godot processes all Bullet collision manifolds as long as the distance is less than or equal to 0 i.e. a collision has occurred. When CCD is enabled, Bullet creates an additional collision manifold with the CCD results. This manifold sets the collision point distance to 0 if a collision occurs. Therefore, when a CCD collision is detected, this manifold is processed too, when it shouldn't.

This PR ensures that Godot doesn't process Bullet collision manifolds' contact points with a distance of 0; so it won't process the additional collision manifold created by CCD.

Fixes #44644.

[AndreaCatania]

In case a collision has 0 distance, we would have both the manifolds excluded. This may lead to some flickering, even if the collisions are still detected. I wonder if exist another way to know if a manifold is generated because of CCD. Do you know?

[madmiraal]

In case a collision has 0 distance, we would have both the manifolds excluded. This may lead to some flickering, even if the collisions are still detected. I wonder if exist another way to know if a manifold is generated because of CCD. Do you know?

The only thing that changes is whether or not a distance of exactly 0.0 is considered a collision and a contact should be reported. There should be no flickering if the distance remains 0.0, because it is never added to the list of contacts. If there is any movement, then it doesn't matter whether or not we consider a distance of exactly 0.0 a collision or not as it will change during the next time step anyway.

[AndreaCatania]

What I mean is that, let's suppose there is a collision and the penetrations between these two objects oscillates between 0.0 - 0.01. The contact report on godot side would appear broken, while Bullet is detecting and keeping that contact. I wonder if exist a better way to detect duplicate manifolds and avoid the above case, if it would became too expensive, it's fine merge this.

[madmiraal]

Bullet requires the penetration depth to be greater than the margin before considering it a collision; so Bullet wouldn't consider a distance of exactly 0.0 to be a collision either.

Do you have a scenario where two objects would oscillate between exactly 0.0 and some small penetration? My expectation is that, if it were to oscillate at all, it would be more likely to oscillate between some number greater that 0.0 and some number less than 0.0, but very rarely be exactly 0.0. Therefore, I wouldn't expect this change to make a noticeable difference.

[AndreaCatania]

Bullet requires the penetration depth to be greater than the margin before considering it a collision; so Bullet wouldn't consider a distance of exactly 0.0 to be a collision either.

Well, it depends on the shapes pair, since each shape margin behave differently.

Do you have a scenario where two objects would oscillate between exactly 0.0 and some small penetration? [...]

From time to time it may become 0.0, but I agree that it's not a big deal lose that info, considering that fixing it otherwise would be too expensive since the distance is the only difference between the standard manifold and the CCD manifold.

[AndreaCatania]

Looks good, Thanks!!
cc @akien-mga

[akien-mga]

Cherry-picked for 3.5.