collection.py: rmap(): Fixed bugs: Made it recursive. Use iters.is_iterable() instead of isinstance(value, list) to work on all iterables. Use value and not nonexistent var list_.

iters.py: Added is_iterable()

parallel.py: prepickle(): Pickle all objects in vars_id_dict_ by ID, not just unpicklable ones. This ensures that a DB connection created in the main process will be shared with subprocesses by reference (id()) instead of by value, so that each process can take advantage of e.g. shared caches in the connection object. Note that this may require some synchronization.

parallel.py: MultiProducerPool.main_loop(): Got rid of no longer correct doc comment

bin/map: Share on_error with the pool

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.

Added dicts.py with id_dict() and MergeDict

Added collection.py with rmap()

db_xml.py: put(): Moved pool.apply_async() from put_child() to put_(), and don't use lambdas because they can't be pickled

parallel.py: MultiProducerPool.apply_async(): Prepickle all function args. Try pickling the args before the queue pickles them, to get better debugging output.