Benchrunner pre-allocates arrays for all iterations which harms performance

[Mikah-Kainen]

Artem

We had a bug before when GHC shared the array value and we sorted a random array only on the first iteration, and all the rest worked on a sorted array. @vidsinghal fixed it with a clever pre allocation strategy. Unfortunately, in my experiments this brings a certain performance cost (probably because GC has more stuff to care about throughout the execution). It'd be good to find a way to avoid that (but not regress back to sharing).

Relevant code is benchOnArrays and its call site.

---

Copied from https://github.com/michaelborkowski/lh-array-sort/issues/41.

[ulysses4ever]

The performance hit (in terms of run time) due to this issue seems negligible, so I pause working on it now. I posted a draft of it in #16. The problem with it is that I split benchrunner into several modules and allocations actually got worse a little bit. So, it needs more work.