Use LikelySubtagsForLanguageV1 for fallback

[robertbastian]

LikelySubtagsForLanguageV1 is almost identical to LocaleFallbackLikelySubtagsV1, the differences being:

• LikelySubtagsForLanguageV1 contains a und field, which is not needed for fallback
• LocaleFallbackLikelySubtagsV1 uses ZeroMap2ds, whereas LikelySubtagsForLanguageV1 uses ZeroMaps with tuple keys. Fallback always looks up by both keys at the same time, so the ZeroMap2d overhead is unnecessary.
• LocaleFallbackLikelySubtagsV1 omits Language->Script entries where the script is Latn. While this might save a few bytes, I'm not sure that it's correct, as fallback for unknown languages will assume that they are Latin and strip the Latn subtag during normalisation.

The ZeroMap2d overhead seems to cancel out the und field and Latn optimisation, and LikelySubtagsForLanguageV1 is only two bytes bigger than LocaleFallbackLikelySubtagsV1 (2230B vs 2228B).

[sffc]

I was 95% sure there was an issue to do this cleanup (including the merging of likely subtags data and fallback data) but I can't find the issue. #3022 alludes to it.