Add way to dynamically add processors to cnlp_processors

[angus-lherrou]

Currently, adding a processor in order to use the cnlp_transformers library for new tasks requires the following steps:

1. Clone the repository and install it in editable mode
2. Write a class in cnlp_processors.py that inherits from the proper base class, defining get_labels and get_one_score
3. Add the class to the cnlp_processors dictionary, defining a task name as the key
4. Add the task name and output mode (mtl, classification, tagging, or relex) to the cnlp_output_modes dictionary
5. If the task isn't a classification task, edit cnlp_compute_metrics to define which metrics to calculate for the task

Steps 2-4 can currently be done dynamically, without editing any files, obviating step 1. Step 5 requires code be edited or the cnlp_compute_metrics function be manually replaced in the cnlpt.cnlp_processors module, which is not a user-friendly workflow.

This process could be simplified so that users never have to clone the repository unless they want to contribute to the project itself; specifically, steps 3-5 could be abstracted into a function so that direct modification of module state is not exposed to the user.

[angus-lherrou]

Ongoing work on #17 and #96 should address this.

[tmills]

this is addressed in v0.4.0