supports mimalloc for ARC/ORC (1.15x ~ 1.22x speedup booting the compiler)

[ringabout]

Credits to @Yardanico
ref https://github.com/Yardanico/mimalloc_nim

I added the commit by
git commit --amend --author="Danil Yarantsev tiberiumk12@gmail.com" -m "supports mimalloc for ARC/ORC"

[ringabout]

The first step is to check whether it reduces booting compilation time. I will refactor it later. Making it work comes first.

[ringabout]

It is weird that importing std/os in the global Nim config increases 200 MiB memory usage.

[ringabout]

The booting time seems to drop at some extent by testing locally and on the CI

ubuntu-20.04 (batch: 0_3) 15.347s => 12.546s （1.22x)
ubuntu-20.04 (batch: 1_3) 9.140s => 7.867s (1.16x)
ubuntu-20.04 (batch: 2_3) 12.374s => 10.688s (1.15x)

[ghost]

I think it's still beneficial to support dynamic mimalloc, because if you have multiple Nim DLLs, or the main Nim program and other Nim modules as shared libraries, if each of them has their own copy of mimalloc, you won't be able to share GC'd objects between the instances. So, similar to how nimrtl has it, we can just allow users to use dynamic mimalloc

[ringabout]

There might be a problem with diagnostic tools like heaptrack and sanitizers that work best with -d:useMalloc. As I understand this switch is now abused to mean mimalloc. Does this PR still allow them to work reliably? Have you tested?

Not really, I add a new --d:useMimAlloc switch. -d:useMalloc still works.

[ringabout]

Waiting for #19972

[Araq]

This is critical for the v2.0 release btw.

[ringabout]

Sure.

[Araq]

Closing for the time being as our allocator is also now threadsafe.