Remove built-in N-Triples parser/serializer about hdt-cpp

[RubenVerborgh]

One usability issue: the built-in N-Triples parser would skip invalid lines. Many RDF dumps are invalid, and SERD just stops when this happens (without clear error message at the moment; would be useful to know which line is invalid). This means you'd have to correct the error and restart generation all over again.

That's not a real reason to keep the built-in parser though, because its definition of "invalid" was too narrow: it would happily parse some invalid IRIs, resulting in an invalid HDT file. At least with SERD, we know that the generated HDT file will be valid.

[artob]

One minor bump in the road of removing the built-in N-Triples parser/serializer is that there are actually call sites in the code base that are hardcoded to use them, instead of cleanly going through a factory function that would encapsulate the multiple parser/serializer implementations.

For example, header parsing in src/header/PlainHeader.cpp is hardcoded to use the built-in parser. I'm working through these now.

[artob]

This is overall a little bit of a larger task than we might have thought, since the varying reader/writer implementations aren't cleanly abstracted nor dispatched through a registry.

I think what I'll do is lift and adapt existing known-good code (that already abstracts over Serd and Raptor, and more) from our datagraph/librdf++ library (used in Dydra), and replace the I/O layer with that. That library also works on FILE handles, meaning that the transparent gzip (etc) compression code could be made to work with it more readily than with C++ iostreams.

[RubenVerborgh]

Seems like a good idea, especially given the FILE handles thing.

[RubenVerborgh]

Removed the built-in parser with bc7a258 and 09a1e6b because people kept hitting this issue (LinkedDataFragments/Client.js#26, LinkedDataFragments/Server.js#37, …).