Array.Parallel.groupBy added

[T-Gro]

This is next part of the RFC to expand the Array.Parallel module, based on fsharp/fslang-suggestions#187

GroupBy Benchmarks were used for picking an appropriate implementation, comparing against:
Sequential Array.groupBy
PLINQ

The tricky part was balancing the optimization with regards to varying settings. Beyond others:
Struct vs Ref
Expensive vs Cheap projection function
Cardinality of the results (# of buckets)

Some properties desired about the results:

• Invokes projection exactly once per input
• Has predictable allocation profile, allocating typically a lot less then PLINQ
• In most of the cases, at least 2x faster than LINQ (for some reasonable configurations, the speed over PLINQ can be up to 5x-7x better)
• If it is slower then PLINQ, it is so only in rarer cases

One of the results contains measurement for the full configuration matrix:
https://github.com/T-Gro/Fsharp.ParallelGroupBy.Benchmarks/blob/main/Results/Detailed_Against_PLINQ_Full_Config_Matrix.md