Fix suspend_process() issue on processes in dirty schedulers

# Context
`erlang:suspend_process()` was broken when applied on processes that
happened to be in a dirty-scheduler. The symptom was:
  - `internal_error` would be returned when called synchronously
  - `not_suspended` would be delivered when called asynchronously,
     even though the process was in fact suspended

# Problem
When trying to suspend a processes in a dirty-scheduler, it is actually
the dirty-scheduler who suspends the process, by calling
`erts_proc_sig_handle_pending_suspend()`. This essentially activates all
pending suspend monitors and finally sends all pending sync messages, by
calling `sync_suspend_reply()`.

The problem is that when calling the latter, it is passing the process
state *before* activating the monitors. In particular, the SUSPENDED
flag is not set, what makes `sync_suspend_reply()` assume that something
is wrong.

Author: jcpetruzza

erts/emulator/beam/erl_proc_sig_queue.c

@@ -5269,6 +5269,8 @@ erts_proc_sig_handle_pending_suspend(Process *c_p)
         msp = next_msp;
     }
 
+    state = erts_atomic32_read_nob(&c_p->state);
+
     sync = psusp->sync;
 
     while (sync) {