fix character controller is_grounded

[ThierryBerger]

• in Make KinematicCharacterController move along obstacles, respect offset more and fix false positives in grounded detection #446 ; there had been a lot of experimentation on that offset. this 1.1 has been kept since the creation of the character controller.
• I think that 1.1 is not enough, I'm not sure exactly why there's number approximation this high, but manual testing results in better behaviour (no red controller from Color visualization for grounded character controller #705), and automatic testing still passes.

Some numbers

offset is CharacterLength::Relative(0.01), then the height of the CharacterController in the exampple is 0.3, resulting in an exxaggerated offset (for aabb computation) of 0.0033. The nudge factor being 0.0001 shouldn´t impact that, but we're still in small numbers game. I think overshooting a bit won´t hurt much, especially if it fixes is_grounded reliability.

TODO?

I think I can make a test about ground prediction (make a character walking on a trimesh for some time, verify it's always grounded), but I fear this test might not always catch such bug.

Alternatives

We can consider exposing this offset multiplier, or the whole AABB used.

[sebcrozet]

It would be interesting to know why 1.1 didn’t work here, but let’s merge this as a stopgap solution. Thanks!

[ThierryBerger]

It would be interesting to know why 1.1 didn’t work here, but let’s merge this as a stopgap solution. Thanks!

@sebcrozet My behaviour analysis, with a capsule moving on a flat ground (character controller example)

It's frustrating that I can´t find why the capsule ends up being too high, but I'll give up on further analysis for now...

https://docs.google.com/presentation/d/1Uc9HEgNH0H3Mf9zvaiuNFCdopPnl50QVsyTkjxdK4po/edit?usp=sharing