Feature/gunney/point search comm filter

[gunney1]

Filter out unproductive communication and searches in DistributedClosestPoint

• This PR is a feature
• It does the following:
  • Implement a way to filter out unproductive communication and searches. This approach doesn't require a user-specified threshold and can work in conjunction with the threshold.
  • Change the correctness for the DistributedClosestPoint so it doesn't require the object mesh to be a complete sphere in 3D. (The complete sphere has polar singularities that are not representative of typical meshes.)

Motivation

The point search currently allows users to specify a threshold beyond which the user doesn't want to look. However, when this threshold is unspecified or large, the implementation can run unproductive searches. It can search and find distant points before searching more closely and discard the previous results. With this change, we examine the bounding boxes of domains before searching. We find a domain that is guaranteed to have points within a computable distance and exclude searching domains that cannot possibly beat that distance.

Approach

The check for unproductive searches is based on the min-max distance of object mesh partitions from a query mesh partition.

1. In the current implementation, we already have all-gathered partitions' bounding boxes.
2. Given any 2 bounding boxes, we can compute the maximum possible distance from a point in one box and a point in the other. For each query partition, find the object partition with the smallest possible maximum distance to it.
3. Use a query-partition-specific search threshold that is the smaller of the min-max distance and the user-specified threshold (if any). If the user provides a very small threshold, this optimization is not likely to help further, but if no threshold is given this optimization can help.

Skipping unproductive searches also removes the associated communications.