The problem is once you let go of those semantics, a lot of software stops working if run against such a "filesystem". If you dilute the meaning of "filesystem" too much, it becomes less useful as a term.

https://en.wikipedia.org/wiki/Andrew_File_System was interesting, I'd actually love to see something similar re-implemented with modern ideas, but it's more of an direct-access archival system than a general-purpose filesystem[1], you can't just put files written by arbitrary software on it. It's a bit like NFS without locks&leases, but even less like a normal filesystem; only really good for files created once that "settle down" into effectively being read-only.

[1]: I wrote https://github.com/bazil/plop that is (unfortunately undocumented) content-addressed immutable file storage over object storage, used in conjunction with a git repo with symlinks to it to manage the "naming layer". See https://bazil.org/doc/ for background, plop is basically a simplification of the ideas to get to working code easier. Site hasn't been updated in almost a decade, wow. It's in everyday use though!