If these really are just graphical variants with no difference in meaning or pronunciation, wouldn’t standardised variation sequences be the way to go? That’s how variant forms in the Myanmar, Phags-Pa, and Manichaean scripts are already being handled, and this case seems very similar to those.