The conversions from julian dates are not necessarily hard, but the I/O functions means having localized names for all days, months, eras of all calendars in all supported languages. If you're thinking of implementing this from scratch (without the ICU dependency), where would these names come from? OTOH if we're using ICU, then why bother reinventing the julian-to-calendars conversions that ICU already does?
i donno why but currently we are using our own function for converting (see j2date and date2j) maybe it's written before ICU but i think ICU helps in adding other calendar support easly. Regarding I/O functions postgresql hard coded days and months names on array and just parse and string compare, if it is not on the list then error(see datetime.c) and it will be the same for other calendar but i think we don't need all that if we use ICU
