Add HDT read/write serialization support

[sihingkk]

This adds HDT (Header Dictionary Triples) as a supported format for Model.read() and Model.write(), using the hdt crate.

What's included

• Reading: m.read("file.hdt") (format inferred from the .hdt extension, or pass format="hdt"). Implemented by iterating the HDT dictionary and mapping into oxrdf terms; literals are parsed via oxrdf's N-Triples-style lexical forms, matching the hdt crate's conventions.
• Writing: m.write("file.hdt", format="hdt"). The HDT four-section dictionary and triple bitmaps are built directly in memory and written straight to the output — no temporary file or intermediate N-Triples serialization. Literals with special characters are stored N-Triples-escaped, following the hdt crate's conventions.
• reads()/writes() reject "hdt" with a clear error since it is a binary format.
• Python tests in py_maplib/tests/test_hdt.py covering round-trips of IRIs, blank nodes, plain/typed/language-tagged literals, special characters, format inference, multi-graph behavior, and the error paths.

Notes and limitations

• The hdt crate loads the entire HDT file into memory on read; there is no streaming.
• HDT is read-only by design, so write always produces a fresh file.
• Cargo.lock grows by the hdt crate's dependency tree.

All py_maplib tests pass (384 passed, 2 skipped).

[thenonameguy]

Writing: m.write("file.hdt", format="hdt"). The hdt crate builds its dictionary from N-Triples input, so writing goes through a temporary N-Triples serialization; literals with special characters are stored N-Triples-escaped, again following the hdt crate.

Please do this without a temporary file/in-memory, so we don't need a 2x memory/disk space for |triples|.

[thenonameguy]

fixes #52