ci: only apply page invariants to real HTML documents

html_extra_path passes files through to the site root verbatim, and search
engine verification tokens are plain text carrying an .html extension. The
glob treated one as a documentation page and demanded a canonical tag,
sidebar nav and three eulerstream links from it, failing the build on a
correct file. Verified the filter excludes only the token: 12 real pages
still checked, and a page losing nav still fails.