Run-time feature detection for AES-NI and TSC

[gnzlbg]

TSC doesn't have an LLVM feature, so IIUC we still can't use simd_test for it (its enabled for all x86 targets). Anyways, offering run-time feature detection for it should do no harm.

[gnzlbg]

I wish we would make the clippy and and rustfmt builds fail hardly. Otherwise every time I do a commit and want to format things appropriately I get a lot of noise in other parts of the library. The same applies to clippy: when I want to check that I haven't written something unnecessarily complex: .map(...).unwrap_or vs .map_or, foo.iter() vs &foo, xxx vs xxx_with (I always forget that the _with methods exist...), unnecessary &foos, unnecessary transmutes, etc., etc., etc. I get a lot of noise from everywhere in the library as well.

[gnzlbg]

So... it seems that the feature cache was full on 32-bit because this PR just added the 32th run-time feature test and I was using an AtomicUsize for the cache...

I have added a debug_assert! for this, which although not perfect, should catch this in the future as long as we test the run-time feature detection system a bit (which we do, against cupid).

I have also increased the size of the cache on 32-bit by making it an AtomicU64. The underlying type of the cache was previously being leaked everywhere, so I've compartmentalized it a bit more, so that in the future increasing the cache again to, for example, be implemented using a pair of AtomicU64s can be done locally.

[alexcrichton]

Hm I think AtomicU64 may not be quite as portable as we might like, but on 32-bit could we use a pair of two AtomicUsize?

[alexcrichton]

(otherwise though this looks good to me!)

[gnzlbg]

@alexcrichton Is there a reason to prefer AtomicUsize over AtomicU32 on 32-bit ?

EDIT: this PR is otherwise ready.

[alexcrichton]

Oh I just tend to prefer stable things to unstable things, but either should work fine!

[gnzlbg]

@alexcrichton we would still the new methods to be const fn for AtomicUsize to work here without unstable features because of the static :/