Fix bounding box leaf collision check

[jorgensd]

Reported by @nate-sime
The following MWE produces the wrong collision for the global_bb_tree

import numpy as np
import dolfinx
from mpi4py import MPI

mesh = dolfinx.mesh.create_unit_square(MPI.COMM_WORLD, 10, 10)

xp = np.array([[100.0, 100.0, 0.0]], dtype=np.double)

tree = dolfinx.geometry.bb_tree(mesh, mesh.topology.dim, padding=1e-4)
global_tree = tree.create_global_tree(mesh.comm)


def pprint(*msg):
    print(f"[{mesh.comm.rank}]: " + " ".join(map(str, msg)), flush=True)


pprint(dolfinx.common.git_commit_hash)
pprint(dolfinx.geometry.compute_collisions_points(tree, xp))
pprint(dolfinx.geometry.compute_collisions_points(global_tree, xp))

giving the following in serial:

[0]: f488f4ecc687d0065315bc9a74939634b18e11ff
[0]: <AdjacencyList> with 1 nodes
  0: []

[0]: <AdjacencyList> with 1 nodes
  0: [0 ]

and the correct result in parallel

[0]: f488f4ecc687d0065315bc9a74939634b18e11ff
[0]: <AdjacencyList> with 1 nodes
  0: []

[0]: <AdjacencyList> with 1 nodes
  0: []

[1]: f488f4ecc687d0065315bc9a74939634b18e11ff
[1]: <AdjacencyList> with 1 nodes
  0: []

[1]: <AdjacencyList> with 1 nodes
  0: []

This is due to a bug in _compute_collisions_point where we don't check if the leaf node actually collides with the point.
@massimiliano-leoni could you see if this impacts your mesh interpolation (in serial)? As the global_bb_tree is used it would add any interpolation point from the other mesh to the process, even if it doesn't collide.

[garth-wells]

Could the PR have a more descriptive title?

And could a test be added?

[jorgensd]

@garth-wells Test added + minor simplification. Feel free to merge this once the CI passes.