Add optional datadict mapping config to AbstractNode

[briangrahamww]

@kzecchini brought up an issue where the sklearn model node requires a hardcoded datadict key that is used by the sklearn preprocessing node, but not necessarily other nodes which may connect to the sklearn model node from upstream. His example is a TF node that creates embeddings, which he wants to cluster using sklearn.

One option is to have a custom key converter / datadict mapping node to rename the datadict key. This seems like overkill for creating a new node.The other option is adding a node config parameter to specify the key to write to the data dict.

We have a lot of redundant code in each of our custom nodes to get upstream data with a specific key and assign it to a variable. We write this code enough that we should maybe abstract this functionality to primrose itself.

One option:

for_example:
  class: ExampleNode
  upstream_data:
    new_data_key_1: upstream_data_dict_key_1
    new_data_key_2: upstream_data_dict_key_2
    ...

would take data_object.data_dict[upstream_data_dict_key_1]
copy it to data_object.data_dict[new_data_key_1]
but then in your code, you still have to use get_upstream_data and then use the new key.

Option 2:

for_example:
  class: ExampleNode
  upstream_data:
    object_1: upstream_data_dict_key_1
    object_2: upstream_data_dict_key_2
    ...

would assign data_object.data_dict[upstream_data_dict_key_1] to object_1

The question is now how does this new object get passed to your custom node like the data_object does?
Maybe add *args to AbstracNode.run?
AbstracNode.run(self, data_object, *args) and methods in AbstracNode that are run on init?

I'm just brainstorming some ideas here. Feel free to weigh in @kzecchini