Don't assume logging is properly set up when handling unexpected errors

This can sometimes cause problems on rubies with real concurrency such
as jruby.  We may try to disconnect before the logger is defined.

lib/new_relic/agent/agent.rb

@@ -521,6 +521,9 @@ def handle_server_connection_problem(error)
           # unknown state
           def handle_other_error(error)
             log.error "Terminating worker loop: #{error.class.name}: #{error.message}\n  #{error.backtrace.join("\n  ")}"
+          rescue # don't assume logger is set up if we're in an unknown state
+            warn "Terminating worker loop: #{error.class.name}: #{error.message}\n  #{error.backtrace.join("\n  ")}"
+          ensure
             disconnect
           end