Hi! I thought I'd take a stab at improving #4966. I've rebased that PR's code upon the current master and made some fixes and changes I've mentioned in comments there.

Differences from #4966:

1. Commits are rebased upon the current master. Merge conflicts (use of items filter) resolved.
2. Updated code to handles cases when there are both sub-documents and links at the same level. I've added a test case for this, with @detail_route().
3. Ensures all HTML IDs only consist of alphanumerics and minus-dashes, so JS won't choke on something unexpected with #some-id lookups (e.g. it had failed on #spam/eggs). I recognize this may be a problem with non-ASCII routes, but I'm not sure how to deal with those. I can think of either using unicode-slugify as an optional dependency, or using hashes instead of human-readable strings. Both options have their downsides, though. Any suggestions?
4. Adds some tests (a bit dirty, as they parse HTML with regexes and assume a particular structure) that ensure links are consistent and snippet code has correct action lists. If desirable, I can probably add a test-time-only BeautifulSoup dependency, and update tests accordingly, so one would be able to update HTML templates without having to worry about tests. But I feel this is probably an overkill.
5. I've updated the code to pass paths as lists rather than strings (prefixes), joining them as necessary. I believe this is a a more flexible approach. This still requires a custom template filter (list_add), but it's a more generic one than split_keys.

I've tested this code in a project I work on, and it seems to work nicely, without any issues. Please review this. I'll be glad to hear any suggestions on possible improvements!

Thanks!