sanity_checks: Also disable builds by CPU family

[blue42u]

In some cases the OS is not enough to determine if a build will succeed or fail, the CPU architecture may also change the result. This PR extends tools/sanity_checks.py to additionally consult the host_machine.cpu_family() (as determined by Meson), if either the system or CPU family is marked false the build is expected to fail, unless the composite {system}-{cpu_family} key is true. By example:

"wrap_name": {
  // Build expected to fail on Linux on any CPU, e.g. Linux + aarch64
  "build_on": {"linux": false},

  // Build expected to fail on x86_64 on any OS, e.g. Windows + x86_64
  "build_on": {"x86_64": false},

  // Build expected to fail on Linux on any CPU, and on x86_64 on any OS
  "build_on": {"linux": false, "x86_64": false},

  // Build expected to fail on Linux on any CPU, and on x86_64 on any OS,
  // except build expected to pass on Linux + x86_64
  "build_on": {"linux": false, "x86_64": false, "linux-x86_64": true},
}

[jpakkane]

Can you post a different example real world project that would need this or is libfpm the only one?

[blue42u]

There are definitely other libraries with architecture requirements, e.g. libunwind or the syscalls Rust crate. Basically any library that uses straight assembly is going to have some restriction here.

That said, of the ones I have interest in developing a wrap for, libpfm is the only one that would fail WrapDB CI without this patch AFAICT. That isn't a very strong argument for this specific feature, I'm open to advice/alternative solutions.