Previously we assumed that ATP-rewritten links would have the querystring immediately after the domain.
This commit updates the regex to match links even if they have a path component.

Fixes #4.

An addons.thunderbird.net reviewer pointed out that innerHTML.replace() is risky, and textContent.replace() would be better. Instead of one big replace on the body contents, we now recursively do replacements on each text node and node attribute.

No support for TB 68 yet, oops.