fix(genai): preserve nested-object required and nullable in tool schemas

[obchain (obchain)]

Why

Fixes #1725. When converting a JSON Schema with nested object properties to Gemini's tool format, _get_properties_from_schema in _function_utils.py corrupts nested objects in two distinct ways:

Bug A — required overwritten. For any nested object property, the source schema's explicit required array was discarded and rebuilt as "every property that lacks a default key". A schema with required: ["name", "role"] and four nullable-optional fields ended up with all six fields marked required. This breaks MCP-style tools and any Pydantic-derived schema with a mix of required/optional nested fields.

Bug B — nullable dropped on second pass. The converter is invoked twice during a single tool conversion: once by _format_json_schema_to_gapic on the outer schema, and again when _dict_to_genai_schema recurses into the inner object. The first pass correctly converts anyOf: [T, {"type":"null"}] → {type: T, nullable: true}. The second pass calls _is_nullable_schema, which only recognises the anyOf shape — not an explicit nullable: true key — so the flag silently disappears.

What

• _get_properties_from_schema: honour the source schema's required list when present; default to [] when absent. JSON Schema and Pydantic's model_json_schema() always set required explicitly, so a missing key correctly means "no required fields" rather than "everything without a default".
• _is_nullable_schema: short-circuit on explicit nullable: true so re-entrant conversion does not drop the flag.
• Four regression tests in tests/unit_tests/test_function_utils.py:
  • test_nested_object_honors_source_required — Bug A happy path.
  • test_nested_object_without_required_defaults_to_empty — Bug A fallback (covers the real production case where a nested object schema has no required key).
  • test_is_nullable_schema_honors_explicit_flag — Bug B unit-level.
  • test_nested_anyof_nullable_preserved_through_reconversion — Bug B end-to-end.

Areas requiring careful review

• The Bug A fix changes the fallback behaviour when a nested object schema has no required key. Previously the converter synthesised one from "properties lacking default"; now a missing required is treated as "no required fields". This matches the JSON Schema spec and how Pydantic / MCP tools emit schemas, but is technically a behaviour change for hand-written dict schemas that relied on the synthesis. I believe the synthesis was buggy heuristic rather than intentional — the outer schema does not use it — but worth a second look.
• The Bug B fix in _is_nullable_schema is additive (one new short-circuit) and should not affect the existing anyOf / type branches.

Disclaimer

AI agents were involved in drafting this contribution; the diagnosis, fix, and tests were reviewed manually before submission.