To be honest, I consider these features to be cheats, similar to save states and slow motion in emulators, so I feel that these shouldn't be available during normal gameplay. With that said, forward frame stepping is already available if you enable cheat mode (enter 0xdeadbeef as a level code, then press C while playing a level): just click any part of the level to advance the frame. And I am willing to add a checkpoint system, though I doubt it would be anywhere near easy.
You have to decide what SuperLemmini is.
Is execution a desirable challenge? Levels become half puzzle, half execution: We shall think about the level first, then shepherd the uncontrollable rodents. Framestepping, savestating, and action-replay are all bad. Remove action-replay to bring SuperLemmini closer to your goal.
Or is puzzle solving the only challenge? Then execution is a hassle, you should minimize it. Action-replay, savestating and framestepping are excellent ways for this. Maybe there are even better ways to remove execution, but we haven't explored them yet.
Some prefer the puzzle-execution mix, some prefer only puzzles. You cannot please both expectations.
After thorough experience in both designs, most players prefer pure puzzles over puzzle-action singleplayer. Few want difficult execution back in Lemmings after playing without. We scratch the action-puzzle itch with multiplayer Lix.
Regarding timed bombers, I don't want to remove them completely, though I'm willing to make them an option to level designers (and even make instant bombers the default). I still want the built-in levels to use them, though.
When you have action-replay or savestates, timed bombers are absurd. Their only purpose is to prolong the playing time. We must implement the same solution idea over and over, until it works. Proxima has laid out the details well.
If you do not agree, please give arguments for timed bombers. I'm willing to dissect the idea further.
I advise against a per-level option for timed bombers yes/no. Players cannot anticipate the program's behavior, instead they must remember details about the inconsistency.
Again: First decide what SuperLemmini is. Then design towards that goal.
-- Simon