Fix non-linux compilation warnings.

[tomusdrw]

I've started adding #[cfg(target_os = "linux")] but for some cases it starts a cascade of related code not being used, so I think it's safer to go with conditionally allowing dead code.

Let me know what's your preference.

[koute]

Hmmm... well, I'm not sure whether gating those to Linux is the way we want to go, due to a few reasons:

• it's not strictly correct (e.g. on Aarch64 Linux systems you'll also get extra warnings)
• we will want to support the recompiler on other OSes in the future too (And in fact the recompiler used to work on macOS too in the past, before I had to refactor a ton of code and didn't have the time to get the generic sandbox updated too.)
• and it seems kinda painful to maintain

So I think what we could do:

• in cases where the code in question is, not sure how to call it, "general purpose", we could unconditionally allow dead_code (for example, Mutex::lock should never give out a warning even when unused, because the whole point of a mutex is locking, so it doesn't make sense for it to not have this method)
• for compiler-specific code wrap it in if_compiler_is_supported! { (which is the proper way to gate code); unfortunately this is sometimes somewhat awkward to do (could be made easier with a procedural macro, but I refuse to bloat up the compile times for everyone just so that it's a little bit more convenient for me)
• Run clippy on the CI twice, but use a target which doesn't have a recompiler the second time to make sure the warnings don't come back

[tomusdrw]

Thanks for the suggestions! I've left a few allow(dead_code) where I felt it's general purpose. The rest was converted to use if_compiler_is_supported.
I've also added a clippy run on macos to prevent new warnings.

[tomusdrw]

Apologies for the borked merge. I run the CI on my fork: FluffyLabs#1 and it passes, so it should be ready for re-review.

[koute]

Thanks!