It's how a lot of newer emoji work.

Suppose, for example, that you want the emoji "Man Facepalming Medium Skin Tone". There is no single Unicode code point for that. Instead, it uses five code points:

    U+1F926 FACE PALM
    U+1F3FD EMOJI MODIFIER FITZPATRICK TYPE-4
    U+200D ZERO WIDTH JOINER
    U+2642 MALE SIGN
    U+FE0F VARIATION SELECTOR-16

The first two compose to make the "facepalm" with desired skin tone. The last two make "man". The zero-width joiner composes the first two and last two together, so that the whole sequence renders as a single emoji.

Here's a list of emoji which use the ZWJ this way:

https://emojipedia.org/emoji-zwj-sequences/

Meaning, does it mess this up?

https://emojipedia.org/zero-width-joiner/

It's a pretty obscure edge case. You could retain ZWJs when they were between two Unicode points with the Symbol or Emoji Symbol classes, but on further reflection, it would probably be more honest to just strip the ZWJs everywhere and decompose the emojis.

EDIT: also what jedanbik said