[Bullet] Seemingly inaccurate kinematic collision remainders in Bullet physics but not GodotPhysics

[BanchouBoo]

Godot version: 3.1.1 Stable 64-bit

Issue description: I made a simple character controller in 3D using a kinematic body and the move_and_slide() method, but when I look at the collision information from the slides, it seems as though a vast majority of the time the collision remainder is a zero-length vector, and if I increase the safe margin it's a zero-length vector 100% of the time. If I swap to GodotPhysics, the remainder is never a zero-length vector, if I increase the safe margin this remains true but it exhibits a bouncing behavior shown below (using a safe margin of 0.1).

Minimal reproduction project: 3D Physics Issues.zip

[AndreaCatania]

A margin of 0.1 is equal to 10cm, try to lower it

[BanchouBoo]

0.1 was just an arbitrary value I picked to test what effect it'd have on the issue, it happens at lower values like 0.05 and 0.01 as well, and even if they aren't values you would typically use they should still work as expected, that is to say they shouldn't cause the bouncing in GodotPhysics, though perhaps I should open that as a second issue instead.

[jitspoe]

I've noticed that, too. It seems if the collision is immediate, the remainder can be 0 and the travel can be negative (opposite of the desired move direction). Seems if the margin pushes something back.

[akien-mga]

Still reproducible in 3.5 RC 1.