AggregateFlows and Benchmarking support for workflows with dependencies

[zeddo123]

This PR makes it possible for the AggregateFlows class (and the benchmarking class) to play well with workflows that have additional dependencies (aka workflows with more than 1 input).

With these proposed changes, the AggregateFlows class can now accept an additional field dependencies that would be used (if not None) when running workflows, thus insuring that they have the correct inputs.

flows = AggregateFlows(input, dependencies=[x, y], workflows=[wf1, wf2])

To add support for the Benchmarking class, the only thing to add is to introduce the class InputWithDeps. When this type is used by the dataset (as the feature), the DefaultAggregateFlowsFactory would automatically "unpack" the object and inject the appropriate inputs to the AggregateFlows objects. Practically, we only need to define our dataset in particular way and octopipes takes care of the rest (of course when the DefaultAggregateFlowsFactory is used).
Here's an example:

class Dataset:
    ....
    def __getitem__(self, index):
         ...
         return InputWithDeps(input=image, dependencies=[bboxes]), ground_truth

fixes #14

[Wala-Touati]

This looks good to me, thanks for the fix