struct InRange: Add type that enforces its value is within a const generic range

[kkysen]

• Part of fn decode_coefs is slow #1180.

Use it, for now, to optimize struct SegmentId (already similarly optimized, but not generically) and fn get_skip_ctx's return value.

Note that we need this for fn get_skip_ctx because it is not inlined, and thus the bounds checks using sctx aren't removed in fn decode_coefs. It is not inlined because fn get_skip_ctx is not BitDepth-generic, but fn decode_coefs is. I think dav1d still had fn get_skip_ctx inlined because it is manually duplicated with the templating system. But if LLVM thinks it is better not to inline this even with an #[inline] or even an #[inline(always)], and we're still able to propagate information through the return type, I'm inclined to leave it un-inlined.