Fix cilksort segfault/speed

[michaelborkowski]

Fixing cilksort segfaults and slowness by porting the optimizations to parallel mergesort (I think it's mainly the recursive helper function and the {-# INLINE #-} )

Haven't seriously benchmarked, but seems slightly faster than parallel mergesort because of the fallback to (seq) quicksort instead of sequential mergesort. On 10 million items:

michael@COS-8MBORK-3912:~/lh-array-sort-newer$ cabal run benchrunner -- 1 "Cilksort" Par 10000000 +RTS -N4
Running OurSort Cilksort (Par)
========================================
array size = 10000000
iter time: 2.155518032
Sorted: OK
BENCHMARK: OurSort Cilksort
RESULT: 10000000
SIZE: 10000000
ITERS: 1
BATCHTIME: 2.155518032
SELFTIMED: 2.155518032

michael@COS-8MBORK-3912:~/lh-array-sort-newer$ cabal run benchrunner -- 1 "Mergesort" Par 10000000 +RTS -N4
Running OurSort Mergesort (Par)
========================================
array size = 10000000
iter time: 3.010259488
Sorted: OK
BENCHMARK: OurSort Mergesort
RESULT: 10000000
SIZE: 10000000
ITERS: 1
BATCHTIME: 3.010259488
SELFTIMED: 3.010259488

[michaelborkowski]

I didn't mean for this PR to be carrying along several old commits... I will update this one later (Don't merge!)

[ulysses4ever]

Thanks for submitting it! A good way to mark a PR as WIP is to use Github's Draft PR feature: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/changing-the-stage-of-a-pull-request#converting-a-pull-request-to-a-draft