Make the package less of a monolith

[parente]

Quick note from the Jupyter call today

• @mpacer mentioned that a simple update to setup.py could make the tool runnable via nbconvert
• I mentioned that breaking up the rather monolithic exporter + preprocessor setup that we have right now to make parts reusable independently (if there's value in that). The gotcha is that REST API calls to Confluence have to be made at various phases of the process to image version lookup, page version lookup, etc.

[parente]

Delayed submit and I crossed the streams a bit with #3. Thanks @mpacer!

[danizen]

The gotcha is that REST API calls to Confluence have to be made at various phases of the process to do image version lookup

This at least is not true in my environment. For issue #30, I simply removed the lookup of the URL and use the attachment filename as in <ri:attachment ri:filename="{{ ... }}"/>. That means that there are parts of the pre-processor that may not be necessary.

Since the other mechanism for doing confluence attachments is needed, I suggest that the template path be also a traitlet, so that people can use different templates for their own confluence.