do we need CollatorType.cldrWithoutFFFx?

[markusicu]

WriteCollationData.getCollator(type) (issue #793 would move this function to class UCA) works with three types, one is cldrWithoutFFFx which builds a CLDR collator except that it leaves U+FFFE and U+FFFF with their DUCET mappings rather than their CLDR tailorings.

Strangely, FractionalUCA.java works with such a collator, even though it writes "SPECIAL MAX/MIN COLLATION ELEMENTS" for these noncharacters, corresponding to the CLDR tailorings.

This type is also used for UCA.Main option testCompatibilityCharacters.

Why? It seems confusing to have this third type, especially to get something different from what we actually output.
Try to remove it and only use either a DUCET collator or a CLDR collator.

If we need and keep this option, then at least consider changing buildCldrCollator(boolean) to buildCldrCollator(enum type) for readability.

@macchiati FYI