Proposal: Obsidian should recognize [latex] … [/latex] blocks, take the content, throw it into some .tex template, run it through a compiler, make it an image, and in preview replace the latex block with the image.
This is basically how Anki handles LaTeX.
The imagefile could be names with some checksum of the latex block to recognize if there already is a corresponding image. Then there would also need to be a way to delete unused latex files, e.g. if one changes something, and so an old image is not implemented anywhere anymore.
The trade-off is that this would take a few moments to do, when switching to preview mode for the first time, and the image has to be generated.
I think that could however be a killer feature, because no web-based competitor is likely to do something like this, and it really makes use of people storing their stuff on their own machine, with their own latex settings. This could easily live besides the current latex support which is faster and completely fine for short things, and if one does not need a lot of custom commands.
If this is worth the effort probably depends on how many people are working with sciency-stuff in the community.