[Discussion] Portability of the libc crate

[marmistrz]

Continuing the discussion from #699.

Summary of symptoms:

• musl in CI cannot be upgraded since an address family was added, on newer glibc tests fail either (issue AF_MAX/PF_MAX incorrect on local system #665)
• newer functions cannot be added without breaking compatibility with the older versions of glibc (see getrandom(2) is missing from the bindings #659)
• a lot of code duplication (the same syscall table in lots of different spaces)

Basically, it's all fun and games while we're adding constants (#define). But when we need to expose a function from libc.so - we're doomed.

I suggested generating the bindings based on what's already installed on the system (e.g. with rust-bindgen)

[gnzlbg]

Duplicate of #423 .