LLVM IR apparently is not restricted to C/C++/Objc/Swift etc, it is a generally purposed IR; and LLVM itself is an infrastructure that contains many facilities to deal with compilation backends (mostly, analyses and transformations).

My understanding is that IR is designed as though there were a VM to run it, but in practice, IR is immediately used to generate code for a target architecture.

That might have been true originally, but I don't think anyone uses LLVM like a JVM/CLR-esque VM any more. As the parent states, the original Low-Level Virtual Machine initialism was even retracted, meaning the project's name is just the "arbitrary" sequence of letters LLVM, with no particular meaning assigned to them.