Handle errors in Process.detach properly

[headius]

The rb_waitpid function in CRuby does not actually trigger an
exception raise when the waitpid errors, leaving that to callers
to decide. Because ours did so, any errors during the
Process.detach call to waitpid would trigger the raise event hook
and unfix #6466.

This change moves the errno check outside of our rb_waitpid
equivalent, making it safe to use in the Process.detach thread.

Fixes #6466 again.

[headius]

An additional change was needed here because even with the previous commit errors may still get raised out of the blocking waitpid call if it is interrupted by another thread. Since the detach thread must be able to interact as a full Ruby thread, we modify this logic to push a dummy frame.

[headius]

The reproduction case is complicated and I am unsure how best to add a spec for it. Basically, the reduced case looks like this:

$ rvm jruby-9.2.14.0 do ruby -rtracer -e "Process.detach(12345).join"
/Users/headius/.rvm/rubies/jruby-9.2.14.0/lib/ruby/stdlib/tracer.rb:134: warning: tracing (e.g. set_trace_func) will not capture all events without --debug flag
warning: thread "Ruby-0-Thread-1: -e:1" terminated with exception (report_on_exception is true):
java.lang.ArrayIndexOutOfBoundsException: Index -1 out of bounds for length 10
	at org.jruby.dist/org.jruby.runtime.ThreadContext.getCurrentFrame(ThreadContext.java:541)

where -rtracer could be replaced by any logic that uses set_trace_func or TracePoint to trace raise events.