File tree diff: normalize URLs before hashing/diff

[stsewd]

What's the problem this feature will solve?

Some tools make use of absolute URLs, including the domain. So, when opening a PR, the docs get build using the preview URL, which is different from the domain used in latest, so all pages are marked as changed.

Describe the solution you'd like

When getting the main content of files, normalize all the URLs so they are relative to the current domain.

[stsewd]

We should also remove comments, or make sure we aren't considering them for the diff.