Should LocaleDirectionality::get perform full Add-Likely-Subtags algorithm?

[anba]

LocaleDirectionality doesn't add likely subtags to und in all cases. Is this expected? I've interpreted Marcus' comment in #3172 (comment) that a missing script subtag is added by performing the Add-Likely-Subtags algorithm and this is also what then got spec'ed into ECMA402's TextDirectionOfLocale.

This test case prints Some(RightToLeft) for und-SA, but None for und-US.

use icu_locale::LocaleDirectionality;
use icu_locale::langid;

fn main() {
  let ld = LocaleDirectionality::new_extended();
  dbg!(ld.get(&langid!("und-SA")));
  dbg!(ld.get(&langid!("und-US")));
}

[anba]

Here's a possibly related issue:

This test prints Some(RightToLeft) for tlh-SA, but None for the other two language identifiers. I'd have expected that it prints None for all three language identifiers, because Klingon in Saudi-Arabia shouldn't suddenly switch text direction.

  let ld = LocaleDirectionality::new_extended();
  
  dbg!(ld.get(&langid!("tlh")));
  dbg!(ld.get(&langid!("tlh-US")));
  dbg!(ld.get(&langid!("tlh-SA")));

[sffc]

Hmm, seems like a bug, because und-US should resolve to Latn? LocaleDirectionality uses locale maximize, but only a partial locale maximize:

icu4x/components/locale/src/expander.rs

Line 537 in ac74590

pub(crate) fn get_likely_script(&self, langid: &LanguageIdentifier) -> Option<Script> {

A unit test should be written and then the algorithm fixed if necessary.

[robertbastian]

The und-US issue is that we don't have likely subtags for und-US, these need to be inferred from the likely subtags for und (en-Latn-US), which maximize does but get_likely_script doesn't.

The tlh issue is that we should not look at the region only if there's a language.