Improve notifications about damaged I/O subsystem

Follow-up to https://github.com/rabbitmq/rabbitmq-server/pull/747

It's the best thing that can be done. Without `exit` anything inside
rabbit could hang - most prone to it is the `start_app` action. Crashing
loudly will be helpful, because during 2016 there were at least 2
production investigations about that issue.

And here is the script that I was using for reproduction:
https://gist.github.com/binarin/edd4102b7262d53e2cf09c76d5ca8e9e

1 files changed, 10 insertions, 8 deletions

diff --git a/src/rabbit_log.erl b/src/rabbit_log.erl
index ed73a293ca..5b5468cab7 100644
--- a/src/rabbit_log.erl
+++ b/src/rabbit_log.erl
@@ -96,7 +96,7 @@ with_local_io(Fun) ->
     Node = node(),
     case node(GL) of
         Node -> Fun();
-        _    -> set_group_leader_to_user(),
+        _    -> set_group_leader_to_user_safely(whereis(user)),
                 try
                     Fun()
                 after
@@ -104,10 +104,12 @@ with_local_io(Fun) ->
                 end
     end.
 
-set_group_leader_to_user() ->
-    case whereis(user) of
-        undefined ->
-            warning("the 'user' I/O process has terminated, some features will fail until Erlang VM is restarted");
-        User ->
-            group_leader(User, self())
-    end.
+set_group_leader_to_user_safely(undefined) ->
+    handle_damaged_io_system();
+set_group_leader_to_user_safely(User) when is_pid(User) ->
+    group_leader(User, self()).
+
+handle_damaged_io_system() ->
+    Msg = "Erlang VM I/O system is damaged, restart needed~n",
+    io:format(standard_error, Msg, []),
+    exit(erlang_vm_restart_needed).