I don’t see any essential differences between the ‘computer era’ and the present. What I do see is an evolution of screen sizes and to a lesser extent input methods and capabilities. TeX on an iPhone is an interesting story. Shortly after the iPhone was released I saw an iPhone running LaTeX. This was written by Jonathan Kew, who also developed XeTeX – a fully Unicode-embracing variation of TeX. It was very fast: if you rotated the phone from vertical to horizontal, the entire file was reparsed in a second or two. Go back to vertical and it is processed again. This never was commercialized, probably for lack of demand. Potential customers probably didn’t see a need to load a TeX file on the phone and then process it, when it was no more work to process the file on the desktop and load the resultant PDF.
BTW, you can change the layout of a LaTeX file by adding one line that loads the geometry package. The dimensions of the PDF output are parameters of the call to load the package. See [https://texdoc.org/serve/geometry/0]
All the problems you mention in your last paragraph, except for one, disappear if you change your mindset to thinking of your document as directory filled with stuff rather than as a text file with pointers (which are invisible until you open the text file) pointing to scattered files.
If you keep MD files and their graphics in a single directory or directory tree, they are easy to move, there’s no way for the graphics to get separated from the text, deleting a document (i.e, the directory holding the parts) always deletes the referenced graphics and only those graphics. Moving to another system such as Obsidian is no problem. The only problem that this method doesn’t address is the case where one graphic file is referenced in two or more documents – shared references. Depending on the size of the shared graphics, you might be better off duplicating the graphics so that each file is referenced by only one text file. Otherwise, you have the same problems you mentioned. (There is a reason that PDF files contains both Objects and XObjects. They are included or referenced in that order.)
There is a slight cognitive load in recognizing that your document is really several file. Perhaps this is the reason the Microsoft Word has Docfiles, TextBundles were invented, Apple invented bundles, Scientific Word used the extension .sci instead of .zip, and why PDF files are hiding the image files unchanged within the structure, after possible compression.
If you have, say, an Obsidian vault, and you want to convert to putting graphics files in a directory shared with their MD partners, and you have kept all the graphics in a single directory, you will have to do some work sorting them out. Finding the backlinks will help. In any case, this is the same amount of work as deciding which graphic files need to be deleted when you delete an MD file, except that you need to do it only once.
I think it is a mistake for each product to have its own solution to problem of aggregating the parts of a document. If they all used zip files, there would be one less roadblock for moving work from one program to another.