Parallelize Phase2 Contributions and Preparation

[gakonst]

• Revives the buffer splitting logic I had added in earlier iterations of the code, to allow for parallel processing of non intersecting parts of the buffer
• Contributions to H and L queries and now done in parallel
• Reading the lagrange coefficients is done in parallel (needed when you create the Phase 2 MPC parameters)
• Creating the lagrange coefficients is done in parallel (needed when you prepare the phase 2 parameters from phase1)

NB: Crossbeam's spawn returns a JoinHandle which we can use to get the values returned from a thread. It's a bit confusing, but Rayon's spawn has no return value. It was also fun to find out that Rust also supported scoped threads, but removed them: rust-lang/rust#24292

This seems to have ~60% speedup at ~200k constraint circuits contributions