Improve TOML lesson data structures

[garrison]

Once the data structures for lessons are documented (#9), the data structures should be improved, and the documentation should be updated in coordination.

Currently, there are three separate lesson interactions:

1. choice
2. storybook
3. podcast

These lesson interactions are unrelated, but they were stored in Ductus using the same data structure, which we called a "flashcard."

Here is a rough plan:

• Identify any lessons that use multiple of the above interaction types. Are multiple interaction types ever used or actually needed?
• Split the lessons into three types rather than having each lesson be type = "lesson" with the actual lesson type under [[interactions]].
• Refactor the "choice" interaction to have multiple groups of cards, rather than dividers. This will be more pleasant and conceptually simpler when editing the TOML.
• Remove headings where it is unnecessary. The role of columns is already hard-coded for storybook lessons, and the best structure for them might be even more straightforward.
• Further improve the lesson TOML for each lesson type, with the goal of making it a pleasure to edit by hand.
• Deal with media elements. We probably don't need both href and blob_href. The lesson elements might not be the best place for credit elements. (Perhaps they can be in a separate section of the TOML, or stored alongside the media elements themselves.)

In the end, we should make sure that the "grid" view continues to work for each lesson type when setting grid = true in a lesson's front matter.

[garrison]

As part of improving the "choice" lesson structure, we should allow an instructor to explicitly specify the "wrong" answers rather than having them come from the correct answers to other questions (see @wikey's #17 (comment)).