Make readme finding of html more reliable

[eteq]

#607 adds some machinery to strip out the html part of the README. However, as https://github.com/spacetelescope/asdf/pull/607#discussion_r233917576 discusses, it may be a bit awkward/annoying/brittle to maintain. So at some point that code should be updated to more cleverly find the html part of the README (see the PR discussion thread).

Once this is resolved it might be this should get pushed upstream to the package-template? But not before this issue is fixed since the only thing making the current system reliable is a specific CI configuration.