feat: add templates_to_graph exposing templates as RDF

[thenonameguy]

Opening PR as a form of technical discussion starter:

• Custom flattened vocabulary instead of standard rOTTR. Chose a denormalized, one-hop-queryable vocab (prefix mtpl, base https://datatreehouse.github.io/maplib/vocab#) rather than the standard OTTR RDF list encoding — optimized for SPARQL analysis and SHACL-shape derivation, not for round-tripping back to OTTR. IMO a custom vocabulary can grow faster over time to support code-generation and validation efforts, especially as different syntaxes and maplib templating options (Facade-X) pop up.
• New namespace https://datatreehouse.github.io/maplib/vocab#. Is this the canonical/permanent base IRI we want to publish under?
• Graph argument defaults to the default graph when omitted — no implicit dedicated graph name. Should we demand users to provide a named graph to keep template metadata separate from instance data? My thought was to enable analysis-only/dataless use-cases.
• Blank nodes are used for parameters/instances/arguments. These get fresh per-call identifiers, so repeated templates_to_graph calls into the same graph will accumulate duplicate blank-node subgraphs. Stable identities would help here, also for tracing lineage for any derivation code.

Main motivating test/high-level entrypoint for review:
https://github.com/DataTreehouse/maplib/pull/53/changes#diff-f1ef54932586718fa9703beb8e469f2cd0c5d054e30790ea3d6cd6deaf8e960aR128

Note, the alternative approach to this template-based codegen is using tools like:
https://pypi.org/project/shexer/