[RFC] ptrace: add generic SET_SYSCALL request

[Roland McGrath]

Not that I'm actually involved any more, but I'd endorse the user_regset
approach and not the new request.  On many (most?) machines, it's already
part of the main integer regset (orig_rax et al) and adding another
mechanism is redundant.  Using user_regset also means there won't be a word
of hidden state that is not visible in core dumps.