My reasoning for it would be that they can be very keyboard specific and might require more normalisation than other character classes.

If I had to make a specific choice I would probably whitelist the most common emojis for some definition and allow those

You typically want a bijection from rendered glyphs to binary representation and restricting to ASCII is the most straight forward way to achieve that.

I think for username it's fine, where a bit of restraint makes sense is for billing/shipping/legal-ish data.

No.

How no? UTF-8 strings has no singular canonical binary representations and typing sequences that correspond to intended texts. Which means it can't be hashed and compared for authentication purposes. No?