deficiencies when changing processed comments

[hendriktews]

I generalized #800 and #803 and came up with 8 tests about changing or undoing changes in processed comments, see #808. Except for one, all are currently failing. If the feature of changing comments without retracting had not been merged, I would vote for not introducing it, at least not in this state. PR #808 adds the following tests. Except 2. they all currently fail.

1. inserting *) inside a comment does not retract
2. inserting *) inside a comment and then some other character does retract - this is the only test that works
3. see issue undo does not retract when undoing a comment created by comment-dwim #800: comment a definition with comment-dwim with comment-style set to 'extra-line, then process, then undo; the undo retracts in Emacs before 27 but not in any later version
4. inside a processed comment, yank something terminating the comment, such as *) XXX (*
5. kill from the middle of one processed comment to the middle of some other processed comment, killing some stuff in between, outside of any comment
6. before processing, insert (* into a first comment and *) into a second comment, such that both comments are fused and some definition becomes part of the fused comment. Then process and undo twice.
7. create a new comment via comment-dwim from the middle of one comment to the middle of some other comment, then process, then undo
8. see issue Task: improve/harden the pg-protected-undo function #803: kill a region from inside one comment to inside a next comment, then process and undo

[monnier]

If the feature of changing comments without retracting had not been merged, I would vote for not introducing it, at least not in this state.

+1