Use distributed data structures for interpolation

[adperezm]

When using the domain_binning option, use distributed data structures.

There are a number of improvements to do:

To make the kd-tree searches more efficient:

• When initializing/creating the global maps: Make the check for GLL points on the bin tree be handled in chunks.
• When Make the check for probes on the Global tree be handled in chunks, just as done for the rank_bbox option.

To distribute the data structure

• Assign bins to ranks in a logical order.
• Each rank should inform the bin owners if they have data in that bin. They do not need to inform that to EVERY other rank as it is done now.
• At rank candidate identification, make the quering rank identify the bins in which it has query points. Then make it ask the bin owner which are the ranks that have points in that bin.

It is possible that for the global structure, instead of using KD-trees, just using a hash table is in fact more efficient. Consider that as well.