What’s the best way to have Rasa simply ignore the difference between it and it's and basically treat them both as the same token? Just add as a synonym?
Everyone uses them wrong anyway, in fact, I’ve noticed the odd phenomena that people tend to use them each for their wrong (opposite) meaning, more consistently than their correct meanings.
This is especially problematic when using pre-trained.
Yes, the reason is that synonyms are for mapping entity values after extraction, to have standardised values when you want to query a database for example. For reference: NLU Training Data
Adding sentence examples to your training data is the right way to resolve this problem though.