Thinking on what causes a disconnect between the framework (sdl2, love2d, monogame) enjoyers and the big-box engine enjoyers, specifically when talking about making your own "engine" is this:
Most people I know rolling their own engine (myself included) do so in a super vertical way, building out just the pieces they need in a style that fits the way they want to program. From the big box users perspective, they see this huge monolith program, impossible to recreate, which is true! But that's not what most of us are doing. When things are written with 1% of the constraints+features, and .00001% of the userbase, they are significantly easier to write and maintain. Moreover, the need for being aesthetically pleasing, or even particularly intuitive to the general user can go out the window, further speeding up time to build.
Additionally, many of the framework enjoyers will not build everything from scratch, they'll grab a collision system from here (box2d), a model importer from here (tinygltf), an ecs from here (tinyecs), and so on, stringing them together under one umbrella.
Platform layer things are also typically handled by frameworks like SDL2/Love2d/FNA, so while an input system is probably being written by these devs, its a much thinner shim library than you might be imagining.
I think using a big-box engine like Unreal is reasonable, so in no way am I telling you not to, this is just trying to lend a little perspective from the other side of the fence and maybe make us seem slightly less insane.
I don’t have any basket in these eggs but I’ve seen enough people wasting time trying to break engines to try to do what they want that I’m a believer. Or at least, yknow, considering using other engines that are less bloated and more specialized in what you’re doing