Break up some of the lengthier complicated sections

[jgabry]

We could simultaneously add more structure to the longer sections and emphasize key points by adding some additional headers (maybe sub-sub-sections without numbers, possibly styled differently than other section styles).

For example, the way the Conditional Probability Distributions section is written it could be naturally divided into

• Definition
• Joint distributions implied by expectations
• Marginalization collapses joint distribution onto component spaces
• Generative modeling is building joint distribution from conditionals

or something along those lines.