Wikimedia production servers are running version 52.1 of the ICU libraries. These libraries support 7 languages that aren't supported by IcuCollation::$tailoringFirstLetters in MediaWiki: dz, ja, ko, my, ps, ur, and zh. We should add tailorings for these 7 languages if possible.
To be clear, when I said complicated, I meant would take me more than 30 seconds to figure out - so it may not be actually that complex.
zh (And probably other ideograph languages like ko and ja) need additional development work because MediaWiki currently makes some simplifying assumptions that aren't true for CJK. Additionally, zh has a political problem of there's not a single collation that's considered acceptable, so we may need to support multiple collations for it to be acceptable to users. @liangent did some work towards that end a while back, but its since bit-rotted a bit.