Convert notebooks directly to pandoc AST

[chrisjsewell]

Rather than using nbconvert (and hence jinja) to convert notebooks, it may be conceptually nicer to convert the notebook into the pandoc AST format, then iteratively apply (panflute) filters to the AST, before output.

Note pandoc (since v2.7) does already have notebook conversion. However, this doesn't handle code outputs very well, and also does not preserve cell and output level metadata.

The conversion should be quite simple as it is effectively a json -> json conversion, we would just need to handle:

1. Specifying an initial preference order for output mimetypes
2. Extracting images etc (from outputs and attachments) to a folder-like object (either on-disk or virtual)