Re-implement `rdf_list` parser without `instaparse`

[inaimathi]

Profiling is pointing to instaparse as the main performance bottleneck. We should re-implement the existing insta grammars using regexes/string slicing directly (there are a lot of assumptions we can make about our input because of the way we're processing it).

This task is for the rdf_list parser specifically. It's lower priority because of how tiny the grammar is (and I suspect that smaller grammars may perform well enough in instaparse, but haven't tested that assumption. It's certainly a lower priority than the core howl and link pieces).