llvm targets on x86 and aarch64

[nia <nia%netbsd.org@localhost>]

currently aarch64 and x86 get all the lang/llvm compilation targets.
no other llvm-supported architectures do.

for aarch64 where bulk builds are already excessively long,
this seems like a bad call in general.

the llvm build for x86 is quite daunting for most people.

cross compilation is a niche thing, and only really AMDGPU is needed by
pkgsrc (to build Mesa).

gentoo is disabling most cross compilation targets by default:
https://www.gentoo.org/support/news-items/2016-10-25-llvm_3_9_with_llvm_targets.html

of those they're enabling by default,

- AMDGPU we probably need.
- NVPTX isn't useful for us since Nouveau apparently doesn't use it.
- BPF ... i'm not sure there's anything in pkgsrc that can take advantage
  of this? the BPF C compiler they reference is apparently Linux-only.

my gut feeling is that all cross-compilation targets should be disabled
on aarch64 to make jmcneill's life easier, and all of them should be
disabled by default on x86 except amdgpu, to make pkgsrc users' lives
easier.

even if you want cross-compilation, you generally don't want /all/ the
cross compilation.