If a reload fails, clear Einhorn::State.reloading_for_preload_upgrade.

If a reload failed, this was causing further SIGALRMs to not bother
reloading, since they thought we were already doing so. In conjunction
with preloading, this meant we would never upgrade our workers.

lib/einhorn/command.rb

@@ -200,6 +200,7 @@ def self.reload
         exec [Einhorn::TransientState.script_name, Einhorn::TransientState.script_name], *(['--with-state-fd', read.fileno.to_s, '--'] + Einhorn::State.cmd)
       rescue SystemCallError => e
         Einhorn.log_error("Could not reload! Attempting to continue. Error was: #{e}")
+        Einhorn::State.reloading_for_preload_upgrade = false
         read.close
       end
     end