WIP: Distributed FMM

[skailasa]

Add an experimental distributed FMM and associated traits for the Laplace kernel.

What is achieved in this PR:

• Significantly clearer trait layout, allowing for maximal code re-use between single and multi-node
• Multiple parallel sorting algorithms: Bucket sort, Sample sort, hyksort
• Ghost data exchange based on neighbourhood communicators
• Builder for multi-node experiments
• Currently only works with unit charges (for benchmarking)

What's not available in this PR, but is of immediate priority (after thesis):

• No load balancing, not really required for uniform data structure
• Currently only contains support for single right hand sides.
• Partial FMM api, no utility functions for clearing and re-attaching particle data
• Helmholtz support
• Python support for distributed stuff

TODO:

• Building on Archer 2
• Building on Kathleen Needlessly antagonising
• check existing Python bindings for single node
• end-to-end tests
• Threading control on a real system.