Consider building/linking `run-make-support` as a dylib instead of a static lib

[Zalathar]

While looking at the code that compiletest uses to build auxiliary crates as dylibs, it occurred to me that run-make tests could potentially benefit from linking to run-make-support dynamically.

This could perhaps result in less bloated rmake executables and faster builds/rebuilds, though that's pure conjecture as I haven't actually tried it. On the other hand, it's possible that the extra effort/complexity required to do so might not be worth it, especially if some of that complexity is in bootstrap.

[Zalathar]

I mostly wanted to make this issue to get the idea out of my head. It's a neat idea, but might not be worth spending time on.

[jieyouxu]

This is possible (though note that some targets don't support dynamic linking, but ig they can fallback to static in that case). run-make-support can also not be built as a stage0 tool lib, it can be built as a regular lib by stage N compiler for --stage N. It was implemented the way it is currently because that was the easiest at the time when I initially added it.