Use radial distance for making LOD decisions.

[TechnoPorg]

Makes LOD distances use radial distance, whereas previously, only the depth forward from the camera was used as the distance. This means that unnecessarily high LOD levels were used for objects located to the side of the camera.

Fixes #54885.

[Calinou]

Remember to check the behavior when using an orthogonal camera. Most importantly, we need to make sure overly aggressive LOD levels are not forced when using an orthogonal camera.

[fire]

I tested on the models I had and it looks ok, but I didn't test the error case #54885

Maybe someone can make a video.

[clayjohn]

Remember to check the behavior when using an orthogonal camera. Most importantly, we need to make sure overly aggressive LOD levels are not forced when using an orthogonal camera.

Indeed, the changes look great for perspective camera, but we may need to have different behaviour for orthogonal cameras like we do for calculating distance for sorting purposes. For reference, this is what we do for object sorting

if (p_cam_orthogonal) {
	depth = near_plane.distance_to(origin);
} else {
	depth = p_cam_transform.origin.distance_to(origin);
}

[TechnoPorg]

Remember to check the behavior when using an orthogonal camera. Most importantly, we need to make sure overly aggressive LOD levels are not forced when using an orthogonal camera.

Orthogonal distance calculation already has different behaviour, as a result of #57419. I'm less certain about how this will work with a camera in frustum mode, though.

[clayjohn]

Looks good to me! Thanks for pointing out that this gets ignored when using orthogonal. As for frustum mode, we don't alter any of our other distance calculations yet, so I am fine leaving this as is.

[akien-mga]

Thanks!

[fire]

Might be a regression? #67479