Draft: Render markdown by converting through asciidoc with pandoc

[gavinwahl]

Use pandoc to convert markdown to asciidoc, then convert the asciidoc to html using asciidoctor.

This has what I consider to be a blocking bug that would only be fixable in the upstream pandoc. Whenever there are multiple instances of ++ on the same line, asciidoc interprets the text in between pairs of ++ to be a "inline passthrough". This makes it hard to use asciidoc to talk about C++, requiring special workarounds.

This mangles, for example, the atomic readme.

Correct, current rendering:

Incorrect, rendering in this PR:

Notice how the link text is not correct and it says C and C11 instead of C++ and C++11.

Here is a minimal repro using the editor at asciidoc.org:

It is probably a bug in pandoc that the ++ are not getting escaped. There are multiple open issues in pandoc around weird asciidoc escaping that seem similar, for example jgm/pandoc#2337. We could consider trying to fix the issue in pandoc, but it'd probably take me at least several days to understand the code.

To see the results on staging, someone will have to clear the rendered content cache. Alternatively, I could change the cache key used for rendered readmes.

Fixes #1328

[vinniefalco]

Can you preprocess the md file and escape the "++" yourself, so pandoc doesn't mess it up?

[gavinwahl]

Not without breaking other things because escaping is context dependent:

C++ should be escaped here in regular text
```
C++ should not be escaped here in a code block
```

The only way to robustly do it is to fully parse the markdown into an abstract syntax tree, which is the approach pandoc takes.

[vinniefalco]

I don't have answers unfortunately...