Handle typing generics in skill API method annotations

[mikejgray]

Description

NeonSkill._register_public_api calls issubclass() on each API method's parameter and return annotations. Typing generics such as Optional[str], List[str], or dict[str, str] are not classes, so issubclass raises TypeError. The error is not caught (only ImportError is), so the whole skill fails to initialize:

neon_skill.py:291  issubclass(signature.return_annotation, BaseModel)
TypeError: issubclass() arg 1 must be a class
ERROR - Skill initialization failed

The fix checks that an annotation is a class before issubclass, for both parameters and return values. Methods typed with pydantic models behave the same as before.

Tests added to tests/neon_skill_tests.py:

• test_typed_api_method: <skill_id>.public_api publishes the request and response JSON schemas, and JSON kwargs from the bus arrive as the request model.
• test_generic_annotations: regression test. A method with Optional[str] parameter and return initializes, publishes no schemas, and is callable. Fails on current dev with the TypeError above; passes with this change.

Issues

None filed. Found while evaluating the typed skill API (as used in skill-date_time) for exposing skill methods as LLM tools.

Other Notes

Locally, 6 existing tests in neon_skill_tests.py fail with OSError: File name too long from combo_lock on macOS with a long checkout path. They fail the same way on clean dev and are unrelated to this change.