Right now, we have to keep the buffer FFI object in sync with the
definition of yp_buffer_t because we access its fields directly. If
we add more fields or change the order, things will get out of sync.
Instead, let's treat yp_buffer_t as an opaque pointer and access
its fields through accessor functions directly. This is more
consistent with how we handle strings anyway.
https://github.com/ruby/yarp/commit/878d845eff