Add ability to override `build_machine.cpu()` in host == build configurations

Currently, we don't have a mechanism to overide the build_machine.cpu() method, like we do with host_machine. This generally doesn't matter, but given that the host_machine.cpu() can only be overwritten when build != host (ie, cross compiling) there's no way to get finer grained cpu constraints in a host == build setup. This could either be done through a build_machine section in the native file, or with a command line argument.

It's important to differentiate this from cross compiling, as it may be important to specify the micro architecture of a build machine, like armv7 vs armv6, which cpu_family() will always return as arm.

See: https://github.com/NixOS/nix/pull/3160 for more details

Start by tracing how host_machine.cpu() is overridden in cross-compiling configurations and how native files or command-line arguments are processed. Compare the proposed build_machine section and command-line approaches, then define tests for overriding build_machine.cpu() when host equals build while preserving the distinction from cross compilation.