The comments make it clear that I didn't want to advocate that this is a huge problem. I see it often enough as a moderator, but it's not my biggest editing complaint.
If this is determined to be a big enough problem, rather than a hard stop, perhaps putting the detected problem edits into a queue for review would be a better solution to this "problem".
A full-on machine filter seems a blunt tool for this ailment and false positives would be damaging to the innocent.
I would bet most of the people prone to making this editorial mistake
- will not see or care if context sensitive help warnings
- won't listen if they do see and understand the warning
- will work around any filters with unicode or worse terms than EDIT!!!
Would asking the reviewers to approve self edits when the first word of a new text added to a paragraph was "edit" or those strings were seen in a title change once edited? Anyone doing human reviewing would be able to clean this sort of thing up / revert the change and it wouldn't need to be an all or nothing, but more like something that could be flagged for low quality if people really are being standard about how they edit a question once it becomes "solved"
I'll leave the determination of existence of an actual problem to someone closer to the data.