Prefer not implementing intrinsics in a way that breaks in edge cases

[bjorn3]

For example saturating_add isn't implemented in a saturating way:

rust-gpu/rustc_codegen_spirv/src/builder/intrinsics.rs

Lines 88 to 97 in f4f1031

	sym::saturating_add => {
	assert_eq!(arg_tys[0], arg_tys[1]);
	match &arg_tys[0].kind() {
	TyKind::Int(_) | TyKind::Uint(_) => {
	self.add(args[0].immediate(), args[1].immediate())
	}
	TyKind::Float(_) => self.fadd(args[0].immediate(), args[1].immediate()),
	other => self.fatal(&format!("Unimplemented intrinsic type: {:#?}", other)),
	}
	}

This is more of a lesson I learnt the hard way than an issue with rust-gpu itself. While it may seem like a good idea at first to get it quickly working, it will backfire in the long term. I have spent quite a lot of time debugging miscompilations in cg_clif caused by exactly this kind of shortcuts. At minimum I think you should add a FIXME every time you add such a shortcut.