fix: populate required fields in FunctionDeclaration json_schema fallback

[giulio-leone]

Summary

Fixes #4798 — required fields lost in FunctionDeclaration when the parameters_json_schema fallback path is used.

Root Cause

from_function_with_options() has two code paths for building function declarations:

1. Primary path (parameters_properties): calls _get_required_fields() ✅
2. Fallback path (parameters_json_schema): triggered when _parse_schema_from_parameter raises ValueError for complex union types (e.g. list[str] | None) — never called _get_required_fields() ❌

Additionally, the fallback path didn't propagate default values from inspect.Parameter to the generated Schema, so _get_required_fields() (which checks schema.default is None) would treat defaulted parameters as required.

Impact

The LLM sees all parameters as optional and may omit required ones. Observed in production: Gemini Flash omitted the mandatory query parameter when calling a tool, because the schema had no required field.

Fix

Three changes to the elif parameters_json_schema branch in from_function_with_options():

Change	Purpose
Added _get_required_fields() call	Populates required field (mirrors primary path)
Propagate non-None defaults to schema.default	Ensures _get_required_fields() correctly excludes defaulted params
Set schema.nullable=True for None-default params	Ensures _get_required_fields() correctly excludes nullable params

Test

Added regression test test_required_fields_set_in_json_schema_fallback that verifies:

• query (no default) → required ✅
• mode (default='default') → not required ✅
• tags: list[str] | None = None → not required ✅

Full test suite: 4726 passed, 0 failures

⚠️ This reopens #4812 which was accidentally closed due to fork deletion.

[rohityan]

/gemini review

[rohityan]

Hi @giulio-leone , Thank you for your contribution! We appreciate you taking the time to submit this pull request. Your PR has been received by the team and is currently under review. We will provide feedback as soon as we have an update to share.

[rohityan]

Hi @GWeale , can you please review this.

[giulio-leone]

Thanks @rohityan. Quick maintainer note on the current head: the effective PR checks are now 14 successful, 0 failing, 1 cancelled (gh pr checks on the current branch state).

The cancelled entry is from the superseded mypy matrix run; the latest mypy-diff reruns are green on the active head, and the full unit-test matrix is green as well.

Happy to adjust anything else if helpful during review.