Reject invalid skill paths

[fallintoplace]

What changed

• build_context now raises a ValueError when skill_path is missing, empty, nonexistent, or points at a file.
• Existing empty directories still scan successfully as valid empty scan targets.
• Tests now cover the graph/API failure mode so invalid skill_path cannot return a low-risk report.
• Graph integration tests set use_llm: False where they only validate report shape.

Why

Direct graph/API callers could pass a typo or invalid skill_path and receive a successful-looking risk_score: 0 / SAFE result. That is risky for CI and automation because bad input can look like a clean scan.

Validation

• uv run --extra dev pytest
• uv run --extra dev pytest -m integration tests/integration/test_graph.py
• uv run --extra dev ruff check src/skillspector/nodes/build_context.py src/skillspector/nodes/resolve_input.py tests/nodes/test_build_context.py tests/integration/test_graph.py