Allow flows to create tasks with custom arguments that survive serialization

A use-case I keep running into is the ability for a flow to define task-specific parameters that will be stored with the task when it is serialized and sent across the wire to workers to process. For example, something like this pseudocode:

flow = Flow('build-cluster')
for node in cluster.nodes:
    flow.add(Task('build-node', node.id))

Since it would be perhaps onerously difficult to support any set of valid python arguments, I would settle for a dictionary, such as:

Task('build-node', task_args={'node_id': node.id})

And then within the Task.execute() we would get the data back as self.task_args['node_id']

The same per-task-instance state is not really accomplishable with the current 'store', and there's no way to scope the store to the task being run, so having the data in the task itself seems like the most appropriate way to go.