CRLF lineendings will happen on Windows,
but they shouldn't matter for Markdown content.
pulldown-cmark will parse them _nearly_ the same.
`<!-- toc -->\r\n` will actually be 2 HTML elements:
the `<!-- toc -->` part and a `\n` part,
whereas `<!-- toc -->\n` will be just one: `<!-- toc -->\n`.
That throws off our marker parser because we're looking for the latter only.
So by stripping our the CR (`\r`) we don't need to special-case anything.
The rendering will be the same.
Closes#35
Roundtripping markdown is actually quite hard.
We don't actually require that.
All we need is once parsing the markdown to find the right marker and
the headings.
We then manually generate markdown and all other content can be copied
unparsed again.
