Move internal fallback option to baked exporter options

[robertbastian]

I also changed the DeduplicationStrategy to default to RetainBaseLanguages, instead of being inferred by supports_builtin_fallback. I haven't change testdata/bakedata generation, as I don't want a data diff in this PR.

[sffc]

LGTM except please change DeduplicationStrategy back to depending on the exporter. I suggest changing the trait function supports_built_in_fallback to instead be default_deduplication_strategy.

[sffc]

Nice!

[robertbastian]

The current architecture raises the question: Are FallbackOptions still fallback options, or are they deduplication options? And do we flatten the options and just use the enum, like with_locales_and_deduplication(..., DeduplicationStrategy)? The enum is non-exhaustive, that might be enough. And do we still need with_locales_no_fallback, or is this now the same as FallbackOptions::no_deduplication()?

[sffc]

The current architecture raises the question: Are FallbackOptions still fallback options, or are they deduplication options? And to flatten the options and just use the enum, like with_locales_and_deduplication(..., DeduplicationStrategy)? The enum is non-exhaustive, that might be enough. And do we still need with_locales_no_fallback, or is this now the same as FallbackOptions::no_deduplication()?

We've gone in circles so many times that I think making things "excessively non-exhaustive" is the right path.

It would be reasonable though to have the following functions:

• with_preresolved_locales(impl Iterator<Item = LanguageIdentifier>)
• with_locale_families_optional_runtime_fallback(impl Iterator<Item = LocaleFamily>)
• with_locale_families_for_runtime_fallback_retain_base_languages(impl Iterator<Item = LocaleFamily>)
• with_locale_families_for_runtime_fallback_maximal_deduplication(impl Iterator<Item = LocaleFamily>)

[robertbastian]

Different PR anyway