When you get more and more stages that are balanced in nature, the pool either widens or stages with previously accepted flaws can become less accepted. P:M seems like a good example of the latter. More isn't necessarily better especially when some stages may have only been accepted as legal for quantity's sake to start with.
I don't think any of the stages in yellow beside Kongo Jungle should be legal.
I have no idea how Stadium 2 is still a debate when three out of four of its transformations are highly disruptive and have the potential to ruin a game. Nobody wants to see a grand finals determined by a mistake made because a player couldn't account for conveyor belts.
I'm afraid Peach's Castle 64 could have issues with people slamming their faces into the triangles and getting stage spiked. If not that, I'm concerned a well-placed tech could save someone's stock by slamming into them. It's the most acceptable one on the list besides Kongo Jungle, but imo it's still a mess.
PS2 really isn't as bad as you think. Ground is only bad if the two players are caught on the complete opposite sides of the stage, but PS1 had the fire transformation, which was just as bad about that, so Ground isn't unworkable.
Ice has two down tilted platforms, and is basically "You get a sliding upsmash, and you get a sliding upsmash, everybody gets a sliding upsmash!" Nothing awful there.
Electric rewards stage control and aerial combat like no other stage I've seen, the belts move so slow a walking Robin can keep up, and IIRC, the stage layout is consistent everytime. It's one of my favorites.
Okay yeah wind sucks, but 3 outta four ain't bad.
Also, if your only problem with Peach's castle is the fact that you can get stage spiked off the steel blocks, then as much as this argument may be dumb, git gud. Teching is not an arcane art here, and hitting the steel blocks is rare back in the 64, so I doubt they're going to be huge factors here.
There are a variety of reasons why all 3 of those are bad stages (you say rewarding player knowledge, but I'm pretty sure people all know how to abuse their properties and the issue is that these properties introduce extreme variance into results).
They do hurt those players because they lead to worse players winning against them due to luck. (castle siege not so much, that stage is pretty yuck but not really banworthy like delfino and halberd)
I suppose this is the main point of disagreement here then. I believe each of those stages has a valuable place in the stage listing. Castle siege has built in anti camping measures for part 2 and parts 1 and 3 are fine stages, Delfino is consistent everytime, so a kill off the top is usually the result of a player with better knowledge getting ready for the change, and taking advantage of the border change to kill. It's common knowledge that it does that, so players should be wary of their positioning before a switch. And Halberd's hazards are slow and predictable, with even the crane being able to be dodged or blocked on reaction. I don't think luck plays a huge factor in these stages, but I suppose we'll simply have to agree to disagree, as the premises of our arguments are not compatible