Rewrite MDX/react component parser into more general-purpose scanner

[NathanLovato]

As part of #59 we'll need to parse all sorts of MDX components. Currently we use a specific regex to parse components instead.

The MDX components part can be done independently from other

There are some little edge cases with the MDX parsing the way we do it, using a regex:

1. It's specific to given components
2. It assumes the components are on a single line

We need to add parsing for more attributes in #63, #64, and #61 to get rid of MDX parser plugins in next, so this part of #59 can be done before the rest of the parser.

I'm thinking of going for a good old scanner and tokenizer with a big switch statement for this, to have the flexibility to gradually add support for all the component syntax (attributes, jsx values with {}, arrays, etc.).

Wishlist

• Work on good error messages from the get-go, like make it explicit which attribute is missing or what exactly is malformed when a component doesn't close properly or has a typo.

[NathanLovato]

Implementation plan:

• Implement through parser Nathan/mdx component parser #71
• Run through document character by character
• Use a regex to find < ... /> or <...>...</...>
• Pass the match to the parser and let it tokenize

This can then replace all specific regexes in the preprocessor for MDX components