People send reactions in response to the meaning of the message content, not to the individual stanza (of which they're largely unaware.)

So, even though the message correction is a new stanza, unless it changes the meaning of the content, their reactions are still valid for that content's meaning.

Obviously it's possible for someone to change the meaning of the content after the fact, but the vast majority of corrections are either fixing typos or adding clarification, and to degrade the user experience by throwing away all previous reactions on a worst-case basis of 'somebody could do bad things' is a poor choice. In the worst case, you'd get reactions that are inappropriate, but it's not quite a disaster (particularly when the edit history is available.)