WIP: Parallel trial scheduler stub using multiprocessing.Pool

[bpkroth]

@jsfreischuetz here's the stub idea I had for making a parallel trial scheduler using multiprocessing.Pool.

This would be implemented in something like a class ParallelTrialScheduler in the mlos_bench.schedulers module.

Perhaps some aspects of the APIs would have to change, and certainly lots and lots of tests and some error handling and interaction with the Storage backend instead of just local dicts for state management.

I think we should probably start with something simple like this in order to complete #380. Maybe for starters it can just handle fanning out the execution of trial repeats across backend workers.

Then later your work for #926 can expand it do budgeted config scheduling and then later model adjusted scoring and unstable config detection.

For testing, I'm thinking we setup something like LocalEnv and just have them run a simple script that returns known data (e.g., maybe even just an echo back the value it's given in the config or a computation of it or some such).

Let me know if you have questions.