Bug: very small vertex input to TriMesh results in panic: unwrap on None when distance_to_local_point is called

[drobnyjt]

Issue occurred after converting vertex input to Angstroms, resulting in very small vertex list:

vertices = [
  [-0.5, -0.5, 0.5],
  [0.5, -0.5, 0.5],
  [-0.5, 0.5, 0.5],
  [0.5, 0.5, 0.5],
  [-0.5, 0.5, -0.5],
  [0.5, 0.5, -0.5],
  [-0.5, -0.5, -0.5],
  [0.5, -0.5, -0.5],
]
indices = [
  [3, 1, 0],
  [2, 3, 0],
  [5, 3, 2],
  [4, 5, 2],
  [7, 5, 4],
  [6, 7, 4],
  [1, 7, 6],
  [0, 1, 6],
  [5, 7, 1],
  [3, 5, 1],
  [2, 0, 6],
  [4, 2, 6]
]

If length_unit below is 1E-10 (1 Angstrom):

let points = input.vertices.iter().map(|p| Point::new(p[0]*length_unit as f32, p[1]*length_unit as f32, p[2]*length_unit as f32)).collect();
let trimesh = TriMesh::new(points, input.indices.clone());

When calling distance_to_local_point(&p, true), parry3d-f32 panics at:

thread 'main' panicked at 'called Option::unwrap() on a None value', C:\Users\Jarat\.cargo\registry\src\github.com-1ecc6299db9ec823\parry3d-0.2.0\src\query\point\point_composite_shape.rs:126:59

Debug output from failing quadtree below: quadtree_debug.txt

[ThierryBerger]

Errors at this scale are expected, that being said, maybe project_local_point_and_get_location would benefit forwarding the Option... it's irrelevant for the majority of cases so it's a bit controversial.