Add more collator data and filtering to testdata; add transliterator attributes domain

[robertbastian]

Currently some of the testdata is only not generated because some files are not downloaded to repo source data. However, this is not what clients can/should do, all of this should be controllable through options on SourceDataProvider, in particular marker attribute filters.

Running make_testdata.rs with SourceDataProvider::new() instead of SourceDataProvider::new_testing() should yield the same results, new_testing is convenient for avoiding the network, but should not behave differently.

[Manishearth]

TIL we filter this for testdata.

observation: we do not filter this for bakeddata, which is the correct behavior

[sffc]

We have filters because it reduces the number of json files.

[robertbastian]

we currently generate the numbering systems for all locales in the data. because we only have select locales in the testdata, this behaves differently between new_testing and new. should we actually filter the numbering systems by locales somehow? the code is:

icu4x/provider/source/src/decimal/mod.rs

Lines 102 to 120 in 8b31bf1

	/// Produce `DataIdentifier`'s for all *used* numbering systems in the form und/<numsys>
	fn iter_ids_for_used_numbers(&self) -> Result<HashSet<DataIdentifierCow<'static>>, DataError> {
	Ok(self
	.cldr()?
	.numbers()
	.list_locales()?
	.flat_map(|locale| {
	self.get_supported_numsys_for_langid(&locale, false)
	.expect("All languages from list_locales should be present")
	.into_iter()
	.map(move |nsname| {
	DataIdentifierBorrowed::for_marker_attributes_and_locale(
	DataMarkerAttributes::try_from_str(&nsname).unwrap(),
	&locale!("und").into(),
	)
	.into_owned()
	})
	})
	.collect())

[sffc]

I don't understand the question?

[sffc]

Question: why did you change the casing? And in make_testdata.rs, you use _. I wasn't sure whether we should use _ or - so I started using _ on my newly added attribute domains since it seems that what we are using elsewhere.

[robertbastian]

make_testdata was an oversight.

we use - in most ICU4X identifiers, I don't think we use _ anywhere. - is easier to type and more pleasing to read

[sffc]

It's not just here, it is other places, too:

https://github.com/search?q=repo%3Aunicode-org%2Ficu4x%20attributes_domain&type=code

I don't want us to become inconsistent. I would rather make a separate PR to change all the instances at the same time, rather than switching just this one. (Please don't change all the others in this PR)

[sffc]

It's not just here, it is other places, too:

https://github.com/search?q=repo%3Aunicode-org%2Ficu4x%20attributes_domain&type=code

I don't want us to become inconsistent. I would rather make a separate PR to change all the instances at the same time, rather than switching just this one. (Please don't change all the others in this PR)