Giving requester chests priority over buffer chests until their requests are fulfilled is the behavior that I thought you were describing when I first read about them. It seems like pretty much an obvious, no-brainer decision to fix it so they do, in fact, work that way. Whether or not you add any extra configuration options, that should at least be the default behavior.
And now on to the subject of belts.
1. When side-loading onto belts was changed to its latest pre-0.16 behavior some time ago -- forcing items into gaps that would otherwise be not quite big enough and pushing back the items behind them -- that behavior was, according to what I'd read at the time, absolutely 100% intentional and not a bug in any way. I clearly remember reading (probably in an old Friday Facts) about how it was being made to work better. Pouring additional items onto a belt like that absolutely should cause items to shuffle around to fit more in if they have room to do so, if only because that's more intuitive in terms of physics, not to mention being very useful in terms of gameplay. (Not to mention that when even mining drills can't load a belt beyond 50-60% capacity without complicated splitter shenanigans, even though they should be literally dropping items on top of the belt where gravity should help them fit in, you know something's really gone wrong.)
2. While I don't believe it was altered in 0.16, I know that side-loading onto the aboveground portion of an underground belt entrance/exit is pretty controversial. As far as I know it was not an intentional feature, but I've never seen a dev call it a bug, either. That capability has existed for a long, long time, and it's a tool that can be incredibly useful in many ways. Saying it "doesn't make sense" is silly. Just for starters, look at the graphics. The hood does not quite extend all the way to the surface of the belt, and you can see the aboveground part of the belt, and the edges of some types of items on the belt, moving into or out of the hood. And even if you discount the graphics by claiming they're "glitchy," the fact is that half of the underground belt entrance/exit is still above-ground -- and inserters can place or pick up items there! If inserters can reach through the solid overhead part of the hood to add or remove items there, then that portion should be capable of accepting items from the sides, as well... Unless you think inserters shouldn't be able to put items there, either (and just look how upset some people are that doing so won't compress belts anymore -- now imagine trying to take that away, too). Finally, considering the fact that there is literally no other way to pull one lane off of a belt (regardless of that lane's compression or mixture of content -- filter inserters are not a viable replacement for this) while completely blocking the other lane, this does not seem like a capability that should be removed, regardless of what the people who dislike it say -- if they don't like it, they can simply continue to not use it, as they always have.
3. On the other hand, simply for the sake of consistency if nothing else, dropping items onto an underground belt entrance/exit should follow the same rules as dropping an item onto a normal belt, regardless of which way the decision goes on inserters compressing belts. If inserters end up being able to compress belts all by themselves (which would be perfectly okay by me), then (and only then) that should occur on underground belt parts as well. And if they can't, then it shouldn't happen on any type of belt, normal or underground. It doesn't make sense for them to be treated any differently (that is, being able to insert even when there are no gaps, as in pre-0.16 behavior) just because of implementation details that should remain hidden from the player. Since inserters have never been able to fully compress a belt by themselves, I have no strong preference on which way that particular decision swings, except that (as I said) it should, for the sake of consistency, be applied equally to all kinds of belt parts -- normal, undergrounds, and even splitters.
In general though, I would strongly suggest that you avoid making belts harder to use or less capable than in previous versions. As others have mentioned, late game tech and "megabase" performance issues alike are already pushing you forcefully enough in the direction of "less belts, more bots." While performance improvements are great, having them come at the cost of huge usability nerfs which end up requiring you to use more belts to make up for those nerfs is just going to chip away at those hard-won performance gains. And it needlessly over-complicates many pretty standard setups, not to mention making some of the more complex use-cases completely impossible.
TLDR: If you don't want inserters to be able to compress a belt by themselves, that's fine (although it would certainly be nice if they could), but combining that with the other usability changes is just way too much nerfing for something that is still in need of buffing to be competitive with logistics bots in late-game.