parallel.py: MultiProducerPool: Pickle objects by ID if they're accessible to the main_loop process. This should allow e.g. DB connections and pools to be pickled, if they were defined in the main process.