fix: frontmatter parsing in windows (CLRF)

[bukowa]

fixes #2628

It's not pretty but it works. ^^

[Gerrit0]

I think we can make this a bit nicer + more robust by using a regex to extract the frontmatter.

If we use firstBlock.text.match(/^(---\r?\n)([\s\S]+)(\r?\n---\r?\n)/), we'll either have a returned object with how much we need to skip when reporting error positions (match[1].length), the yaml text to parse (match[2].length) and the new start position of the first block (match[0].length), or no object, in which case there wasn't frontmatter. Thoughts?

This will also handle the issue where a file has mixed line endings, as some people don't normalize those...

[Gerrit0]

I spent a bit more time looking at this... and turns out that earlier assumption was bad because of eemeli/yaml#127, I ended up pulling your branch to play around with how to fix that, and fixed it another way..

[Gerrit0]

I wasn't thinking, so rebased when looking at this locally instead of merging... merged with:

21cf5ae
ef0f26e