It now also removes the va_list functions as all of them can be replaced by ... functions, and va_list seems to behave very differently between architectures and is not really usable in rust yet (this is all just from my understanding and probably wrong).

Sorry for only noticing this only after merging and looking at the main branch CI logs.

The windows build now prints warnings about u128 being not being FFI-safe. This is because fz_context.error.stack[0].buffer is of type jmp_buf which has an alignment of 16 on windows. To keep this alignment bindgen uses u128 inside the opaque type.

This warning is, however, is not releavant for us for many reasons:

1. we never pass a fz_context over FFI boundaries, only *mut fz_context
2. this has been fixed in rust 1.86.0 ...
3. ... and therefore the warning is planned to be removed as well

What I think would reasonable is to either

• allow(improper_ctypes_definitions) in mupdf-sys. This would silence other improper_ctypes_definitions warning that might be useful though.
• skip opaque-ing fz_context for now, hiding the types with this PR is more of a "style" and "correctness" thing, so removing it for now would be not a problem.
• adding fz_context as our own opaque type not generated by bindgen. As we never need to create a fz_context on the rust side, but only have a pointer to it given to us by mupdf, we don't have to worry about it's alignment, and therefore don't have to fill it with u128 to achieve this. This sounds more confident than I am to have gotten the details right, but I think it makes sense.

Feel free to send follow-up PRs and merge when CI passes.