Support slicing likely subtags and adding extended likely subtags

[sffc]

CLDR is planning to expand the number of entries in likely subtags data from the current ~1000 to ~7000. This isn't sustainable with our current use of likely subtags for locale fallback.

I suggest we take an approach similar to the one we did with Japanese eras: the default key is defined to contain only the essentials, and an additional "extended" key contains the full set. The choice of which one you want is done at the constructor level.

Based on a discussion with @macchiati, CLDR can add some spec text to inform ICU4X of how to correctly slice the likely subtags data.

I would also like to use this opportunity to possibly consolidate the two copies of likely subtags: the one for fallback only, and the one for the LocaleExpander.

CC @zbraniecki @dminor who have worked on this component.

[sffc]

https://unicode-org.atlassian.net/browse/CLDR-15071

[sffc]

Based on the discussion in #3022, we should instead favor an approach more like Normalizer, where the extended subtags are a supplemental key that sits on top of the core key without duplicating the data in the core key.

[sffc]

I would like to end up with three data keys:

1. Likely subtags needed for fallback, in core locales only:
   • L → R + S
   • L + R → S
   • L + S → R
2. Remainder of likely subtags, also in core locales only:
   • S → L + R
   • S + R → L
   • R → L + S
3. Extended data with non-core locales and all six mapping types

In addition, I would like to take some low-hanging fruit to reduce the data size of each of these bundles. I opened https://unicode-org.atlassian.net/browse/CLDR-16427 with one thing that should help a lot.