Reasoner-transpiler

This version was mostly more refactoring of attribute handling for performance. It also moved knowledge source / provenance handling inside of transform_attributes so that Plater or other applications could call transform_attributes on a node or edge and have it convert every attribute into TRAPI without needing additional manipulation.

This fixes a couple issues introduced by recent changes:
- some attributes were being included twice for edges in the knowledge_graph
- subclass edges that are returned in results as part of a variable length subclass edges and as a normal edge in a result could produce unintended or inefficient behavior

This improves handling of edge binding ids when they don't exist in the graph, making them "edge_1" "edge_2" etc

This release adds support for handling async neo4j driver responses back into transform_result, and an async driver to the pytest fixtures for testing.

1. Reworked subclass handling. No longer return subclass node ids with a query_id in node bindings. Instead, when a subclass edge and node are used to infer a result, always return the queried node ids as node bindings, and create new edges that are composites of edges that match query edges and the subclass edges that were used to infer that result. Use support graphs (auxiliary graphs) to show when these edges were used to generate a result. Queries without edges do not perform node subclass inference anymore.

2. Updated dependencies, actions, and tests to support neo4j 5 and python 3.12 (as well as python 3.9).

3. Removed HTTP protocol functionality and NEO4J_PROTOCOL env var usage.

4. Added a fallback for attribute mapping lookups. If an attribute is not included in explicitly defined attribute type mappings, use the biolink model to see if it's valid biolink.