fix: honor required fields in JSON schemas

[yaodong-shen]

Summary

• honor each object schema's required array when creating Pydantic fields
• use declared defaults (or an omittable None default) for non-required properties without making explicit null valid
• preserve the existing contract that all skill response fields are required
• cover top-level and nested required semantics with regression tests

Fixes #833.

Validation

• black --check adala/utils/pydantic_generator.py tests/test_pydantic_generator.py
• python -m py_compile adala/utils/pydantic_generator.py tests/test_pydantic_generator.py
• pytest -q tests/test_pydantic_generator.py tests/test_skills.py tests/test_label_studio_skill.py tests/test_analysis_skill.py (22 passed, 3 skipped)
• full pytest: 78 passed, 4 skipped, 9 deselected; one unrelated Gemini image VCR test fails because the response lacks content after its image fetch path. The same failure reproduces on clean origin/master.

[yaodong-shen]

The Check, pytests, Docs, and CI/CD Pipeline workflows are waiting for maintainer approval and created no jobs for this external-fork PR. The focused and full local test suites pass apart from the documented unrelated Gemini baseline failure. Could a maintainer approve the workflow runs when convenient?