Remove <base> tag, use hrefs prepended with base

[jdkoren]

For technical reasons, we can't really compute and inline the href base when accessing a model's href, so instead we use a placeholder and substitute it just after rendering each template.

See discussion in #2090

[jdkoren]

This was just off the top of my head, happy to replace it with something else.
The backslash escapes are unnecessary, but should prevent this string from being substituted if someone runs Dartdoc on Dartdoc itself.

[jcollins-g]

This string is short enough I can imagine someone else using it reasonably. Maybe: '\%\%__HTMLBASE_dartdoc_internal_only\%\%' instead?

[jcollins-g]

This in general looks OK.

I'd like to see that a test verifies behavior with manually constructed links to elements using the old and new assumptions.

[jdkoren]

This in general looks OK.

I'd like to see that a test verifies behavior with manually constructed links to elements using the old and new assumptions.

I added two tests to model_test, but they only check that the relative hrefs in the doc comment is left unchanged; there isn't really a way to verify that one of those is actually a broken link (Dartdoc does emit a brokenLink warning for it though). Let me know if you have other thoughts on what we could look for here.

[jcollins-g]

Alright, this is looking doable, even if the implementation is more than a bit horrifying in practice. I would like to see this behavior controllable via a flag (OK with default-on, we'll just shut it off anywhere it causes a problem if need be until we can fix downstream users). I also want a big TODO to lock down TemplateData as I discussed in the original bug. But I'm otherwise willing to tolerate it in the meantime.

[jcollins-g]

This string is short enough I can imagine someone else using it reasonably. Maybe: '\%\%__HTMLBASE_dartdoc_internal_only\%\%' instead?