Check out this test site with both tabs and footnotes with same ids working correctly:
https://gist.github.com/6d82074d3539851e6eaec6dc8db72dd2

This also fixes the issue I had mentioned where same headings in different docstrings would share the same id.

#a.foo, #examples, #b.foo, #examples becomes
#a.foo, #a.foo--examples, #b.foo, #b.foo--examples

View test site: http://pryp.in/p/6d82074d3539851e6eaec6dc8db72dd2

Once again, fantastic contribution @oprypin 🚀 Thanks!
LGTM, lets see how it behaves in the regressions workflow 🙂

I notice that with this change, it is now possible link to stuff like [foo][some.identifier--subdoc-heading], and people are figuring it out. Should we embrace this or lock it down?

I think we should embrace it, this is a pretty nice feature. What's your opinion?

Overall I support it, I'm just disgruntled by the fact that I didn't realize this effect when I developed this PR, and also the choice of the "--" separator wasn't with this usage in mind, although maybe there isn't anything better.

For example, maybe people would've wanted to use some.identifier#subdoc-heading - but on the other hand, I'm already using the # symbol for identifiers in Crystal language, so it'd be ambiguous.

I like --. Headings IDs get slugified with - as separators, so a double -- makes sense to me to delimit the object identifier and the subheading. A bit like Django's ORM parameters using double underscores to avoid mistaking for another attribute: attribute__contains="some string". Or like escaping " with "" in CSV.