If part of a heading is bold in a Google doc, the bold text does not get surrounded with ** bold markdown sytnax — except in Firefox.

For example, in this document: https://docs.google.com/document/d/1lU6N1g5jmyvxZebY1JZpdVtS5SOLlaf65i1b7FilRfY/

Results in the following markdown on Safari and Chrome:

# Heading with bold and _emphasized_ text

Normal text

In Firefox, the bold is preserved, and you get this output:

# Heading with **bold** and _emphasized_ text

Normal text

I think what we’re gettin in Firefox is better, and we should make sure this happens in other browsers, too. (Even though Markdown is really a shortcut syntax for HTML and most HTML renderers will already make the heading bold, there is semantic meaning here worth preserving. And some Markdown renderers might handle this case specially and still differentiate the text from the rest of the heading somehow.)

As far as why this happens: Google Docs wraps the content of headings in <span style="font-weight: 400"> (for normal text) and in <span style="font-weight: 700"> (for explicitly bolded text) the same as anywhere else in a document. But some browsers (WebKit and Chrome, but not Firefox; I’m not sure about other more niche engines) normalize the styling of the bolded text by removing the font-weight property, since the weight can be inherited from the heading it’s in (since the browser’s default styling of the heading is to make it bold).

I can see two obvious approaches to solving this:

1. Identify headers where some text has font-weight: 400 or font-weight: normal and other text has no font-weight at all, and update the latter text to be bold.
2. When resolving styles in lib/css.js, include some default styles for some elements (in this case font-weight: bold for headings).

The second approach is probably simpler, but it might result in unnecessarily bolding text in headings that are pasted in from other web pages, e.g:

# **Heading Text**

While non-Google-Docs source HTML is not our primary concern here, it would be nice not to make it worse than the current situation.

Note: this was originally mixed in a bit subtly with the tests in #104; I’m extracting it into a separate issue here.