Handle the sr-Cyrl-ME -> sr-ME collation-only fallback

[robertbastian]

Fixes #3287

This handles the sr-Cyrl-ME -> sr-ME fallback the same way we handle the other collation-only fallbacks: by explicitly adding the data for sr-Cyrl-ME so we don't go through the default fallback mechanism for that locale.

Changelog

N/A

[sffc]

Would be nice for the test to actually run the Collator constructor but it is a bit tricky... good for a follow-up.

[hsivonen]

Do I understand correctly that explicitly asking for Cyrillic collation for Montenegrin now results in the Latin collation similarly to how asking for Cyrillic collation for Croatian results in the Latin collation ... but unlike Bosnian and Serbian, which do allow explicit request for the non-default script.

Why? What's the upstream CLDR issue motivating this change?

For reference, here's the Firefox/SpiderMonkey test case that documents the situation before this PR: https://searchfox.org/firefox-main/rev/23974e2d947e31e4ae42ae2758a4416c9a6d8671/js/src/tests/non262/Intl/Collator/bcms.js

Also, AFAICT, there is no technical reason why we couldn't merge the Latin and Cyrillic collation data for Bosnian-Croatian-Montenegrin-Serbian and make Latn vs. Cyrl a matter of script reordering on top.

Are users of Bosnian-Croatian-Montenegrin-Serbian actually better served by having the other script collate according to root as opposed to having the other script also collate according to language-specific rules?

[robertbastian]

Apparently these come from upstreaming ICU behaviour to CLDR: unicode-org/cldr#2664, unicode-org/cldr#3504.

I believe it was initially added to ICU in icu4c/source/data/icu-coll-deprecates.xml, which was commited as "Merge CLDR25 data into trunk". However, I cannot find any reference to sr_Cyrl_ME in CLDR 25, so I believe that file was handwritten. sr_ME is listed with the other sr variants there (and other multi-script languages), and it's the only one where the added script tag is not the likely one¹. It looks suspiciously like a typo.

The initial aliases from that file have since evolved through

• tools/cldr/cldr-to-icu/src/main/java/org/unicode/icu/tool/cldrtoicu/IcuConverterConfig.java
• tools/cldr/cldr-to-icu/build-icu-data.xml
• tools/cldr/cldr-to-icu/config.xml

and the ones that are just likely subtags have disappeared, leaving just sr-Cyrl-ME -> sr-ME. Along the way, helpful comments like

It is not at all clear why this is being done (we expect "sr_Latn_ME" normally).

have been added and removed. It still says

TODO: Find out and document this properly

today, but that work is not being tracked anywhere, and apparently wasn't enough to have someone look at this before upstreaming it into CLDR.

Footnotes

1. Note that Cyrl was the likely script for sr-ME until CLDR-2203, but that was way before CLDR 25 ↩