HTML provides two general elements, where div is a natural block element, and span is a natural inline element. All other elements are similarly assigned to be a natural block or inline.
Now, while both can be made by css display to be any of inline, inline-block or block, they are still treated for enclosure purposes as their natural selves, hence the warning messages. Leopards and spots sort of thing.
However, css is only meant to be for making what an element looks like (presentation), but not actually be like (functionality), so it doesn't change an element's basic nature, though that gets very fuzzy in practice. A span made block becomes a bully that kicks everything else off the line, which is very un-inline sort of behaviour.
So, to mitigate against possible conflicts between their natural and css-induced behaviours, it is better to allow:
div or any natural block tag to only ever be block or inline-block.
span or any natural inline tag to only ever be inline or inline-block.
This will also mitigate against tending to build page structures that will likely end up churning out error and warning messages.
Basically, NEVER embed a natural block tag inside a natural inline tag, at any depth.
Why there is a really a distinction is perhaps due to a simplistic idea of what HTML was going to be used for when it was first dreamed up.
Certainly, framework makers got around a lot of these what-to-embed-where problems by just using myriads of divs everywhere, and 'divitis' was born, and still alive and well in every framework. Just have to press F12 in a browser on almost any commercial web page and drill down through a dozen divs. This very page has 15 unbroken levels of divs.
It is not hard to see why just settling on divs made sense. For example, a p tag may have a bunch of links to various sites, and that is ok because inline links are allowed in a block p. However, if not wanting to have query variables visible in those urls, then buttons are required. If only one, then the p can be put inside a form, as a p cannot contain a form.
The formaction attribute on a button can be used to target a url other than the form default, but it still does not allow independent forms, each with their own set of hidden inputs. A button can use the form attribute to use it with a form that isn't an ancestor, but it can get messy to keep track of.
For multiple links to different sites to appear as part of one paragraph though, the only way is to use a div instead of the p and then wrap each button in its own form set to inline. Most frameworks have to cope with so much more complex scenarios that nested divs are the only way to go.
It meant that they really only had to manage one tag per purpose and manage it as if it was an isolated environment. So what was meant to be an occasionally-used functional grouping tag became the web's Lego block. And none of them are going to risk breaking their frameworks by converting to HTML5 semantic tags in a hurry. In the end, semantic tags only really work for fairly static content rather than rich interactive sites.