1) Why do we need this mysterious machine level? If the machines could run the system on their own, why develop the Matoran? If that GBs intended the system to be run by the Matoran, to avoid Baterra-style robot problems, why make the machine level?
Because they wanted near-sapient beings for their creativity in dealing with unforeseen problems, but wouldn't want the everyday repairs to be done by such beings who might misapply that talent and cause other problems.
Keep in mind that we already know the Matoran don't directly repair things beyond the domes. We thus need something like a "mysterious level" to then explain how their work equates to repair of those things, whether it's machines or something else. The option of "Matoran travel anywhere needed to fix things" isn't given to us by the story team.
So, however the mysterious level works, clearly the GBs intended the system to be run by Matoran, and this is simply part of that system.
It's like, using a computer like I am now; I'm just doing a little bit of work, and a bunch of tech translates that little bit into comparatively big accomplishments (turning it into binary code, electrical signals, wireless signals, connecting to the internet, sending the message around the entire globe, etc.). Likewise, the Matoran do a little bit of work, putting replacement parts as instructed into automated towers, and systems move those parts wherever needed, and install them. By the rule of Ockham's Razor, seems to be the simplest explanation.
2) You seem to imply that are automated robots that make repairs to the robot's systems, and that the Matoran repair these. Why would they not repair these systems directly?
Because then they'd need to know about them. They'd need to know about the 90 something percent of the machinery around them that aren't contained on small spots of islands or continents which themselves are just little spots on the insides of the dome-shaped spaces of air. We don't know how much of what surrounds them is just rock, and how much essential, vast machinery, but it seems to be a good bet that there is a lot of huge machinery out there, and since the Matoran only know of what little is on the lands, they can't be the ones doing those repairs. There were even Rahi to actually prevent them from going near these systems.
In other words, it's a simple deduction from the fact that your essential repair workers (whether just "partially sapient" as Greg confirmed the GBs knew the Matoran were, or fully as they turned out to be) are kept both physically away, and ignorant of, most of what they need to keep functioning, and yet just not working for a while during the civil war caused Mata Nui to nearly die.
It wasn't that Matoran weren't taking trips to spots deep underground or in the ceiling of the domes to make repairs -- just not keeping the automated skyscrapers in Metru Nui going was sufficient for Mata Nui to be dying. This implies there's some kind of technology that "reaches" from these skyscrapers to essential systems. Whether it's mobile robots as I theorize or something else, it does seem there has to be something. And it can't be just Mata Nui's power because he doesn't pay attention to his insides.
The only other option, it seems to me, is that his most essential functions are entirely contained within those automated skyscrapers themselves, which seems unlikely for a robot of this size. Although I suspect the truth is somewhere in between these two; some key functions surely are in the skyscrapers, but also something has to explain how Matoran work in a few small places (compared to size of robot) achieves repairs throughout the entire robot.
3) Why would the GBs value that the fact that the Matoran are actually maintaining a giant robot to be a secret (as a possible solution to number 2), if the Matoran were not meant to be sapient in the first place?
This is where I think it's key that Greg said they did know they were partially sapient. They apparently didn't expect culture, or evil, but they might have foreseen things like beings with such intelligence and imagination having a stronger potential for losing mental grip on the mundane things. The very things needed to solve unforeseen problems could make them have trouble coping with mundane, easily-foreseen problems that are the majority of needed repairs. Something like getting bored and snapping, perhaps, and realizing the scope of what they were keeping running could then be a really bad thing.
Say, one decides everything is just too boring and it wants to end the whole thing. If he just thinks he's on a mound of dirt in an ocean under a sky of stars, keeping some mysterious gizmos in a city that are strictly conveniences for some complex city tech, he won't think that sabotaging that tech is likely to make everything end. Probably the glitching unit would just commit AI suicide. But if he knows how to bring the whole place down -- if everybody does, and they're all prone to getting bored... the risk of it getting brought down sooner or later in such a way goes up. Or maybe some "non-evil but similar enough to be a threat" version of raw ambition could lead one to do something like Teridax's takeover.
Keep in mind too that Angonce's reaction implies that it's not like the GBs thought they could make sapience and "didn't want to" make them that way. It's that they never dreamed they could achieve it. So, sapience does seem to be roughly what they aimed for, they just assumed it wasn't true sapience and was just a best-guess AI approximation. But even that close approximation could come with many problems. They seem to have intended emotions, for example. The everyday repair tech wouldn't, so it wouldn't have a problem doing the same basic job over and over and over. But that's all it could do, so if it runs into a new problem it isn't programmed to handle, it can't fix it, so it needs to go to a "reservoir of creativity" -- the beings contained in the artificial environment.
Another thing is just quality of life.
The GBs might go to drastic measures sometimes like the Baterra, but they seem to be motivated out of empathy for living things. Even as "not quite sapient", they would still be creating something like really smart animals, and it seems more humane to have them do only a little work in locations designed to appeal to them (plants, animal life, starry skies, cool breezes, etc. -- not ugly metal walls only), and have tech "extend the reach" of their work. And keeping them ignorant of exactly how their work is important may itself be more humane too. Believing yourself to live on a self-stable planet is probably a lot less stressful than knowing you're inside a shell of metal careening through the deep cold nothingness of space. (And then being less stressed probably makes you a more efficient worker too.)
So, in short, probably the simplest answer is, the theoretical repair robots have no emotions, thus no problems caused by emotions, yet also thus no imagination. The Matoran are the inverse -- to have imagination to solve unforeseen problems, they had to be emotional by design, but thus have problems caused by emotion, whether as "smart animals" or "fully sapient like humans." By having both in the system, theoretically any problem could be solved.
Checks and balances.
BTW, we could apply the same question to the Bohrok -- why not have Matoran or Toa travel outside the robot to clean off the camouflage island? They'd be physically just as capable. But then, again, they'd have to know about it. So, since robots were used by the GBs for that problem, it seems natural to me to theorize that robots also probably solve the other. Just probably way less of them, since repairs are needed a lot less often than clearing the camouflage (needed every single time he leaves a suitable oceanic planet, at least).