[dundysm]

fixes #461

what

Bvh::leaf_pairs seeded its traversal with the root-level node pairs (next plus up to three stack entries) without calling check on them. LeafPairs::next returns any leaf/leaf pair it pops, so for trees of one or two leaves, or trees whose root has a leaf child, pairs that check rejects were yielded anyway. with the report's two disjoint trees it returned [(0, 0), (1, 0), (1, 1), (0, 1)] instead of [].

this runs check on every seeded pair before it is queued, the same way Leaves::new already does for the single-tree iterator (#372). the seeded next also requires both left roots to hold a leaf, so an empty root never comes out as a pair.

tests

• new test_leaf_pairs_checks_root_pairs in bvh_tests.rs: disjoint trees of 1, 2 and 3 leaves yield nothing, and partially overlapping trees (1x1 up to 16x9 leaves) match a brute-force aabb check. it fails on master (1 leaves: unexpected pairs [(0, 0)]) and passes with the fix.
• cargo test -p parry3d --lib (81 passed), cargo test -p parry2d --lib (98 passed), cargo test -p parry2d-f64 --lib bvh (5 passed).

[dundysm]

@sebcrozet small fix for #461; leaf_pairs now runs check on the seeded root-level pairs like Leaves::new does, with a regression test that fails on master. would appreciate a review when you have a moment.