Unify back end dispatch and precompilation effort

[williamfgc]

This issue tracks progress towards unifying the back end dispatch and allowing precompilation. This is the outcome requirements from the agreements in our offline discussion on JACC internals.

Tasks/goals:

• Common solution: add dispatch argument to parallel_for, parallel_reduce private API (e.g. PIMPL)
• Keep naming simple and keep the public API the same
• Moving Array global to a struct variable. Lifetime needs to be understood for this struct
• Have at least one MPI.jl test to test JACC.Array interoperability under this new design
• Test this new redesign on HPC systems: e.g. Frontier, Perlmutter, Aurora (TBD)

Unifies PRs #118 #114 #86 #62 and resolve issue #53 #98
Note: @PhilipFackler will be the lead developer and maintainer.