fix: handle numpy scalar types in safe_json_value

[avasis-ai]

Summary

Add explicit handling for np.integer and np.floating scalar types in safe_json_value().

Problem

safe_json_value() handles np.ndarray (via .tolist()) but misses numpy scalar types like np.int64, np.float64, etc. These fall through to the else clause and are returned as-is. Since json.dumps() cannot serialize numpy scalar types, this causes a TypeError at runtime when tool results contain numpy scalars.

In numpy >= 2.0, scalar types no longer inherit from Python builtins, making this even more likely to fail.

Change

Added two checks before the existing np.ndarray handling:

• np.integer → int(v)
• np.floating → float(v)

Related to #34405 (parameter precision loss with numeric types).

[fatelei]

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
================================ tests coverage ================================
_______________ coverage: platform linux, python 3.12.3-final-0 ________________

Coverage XML written to file coverage.xml
Coverage JSON written to file coverage.json
=========================== short test summary info ============================
FAILED api/tests/unit_tests/core/mcp/session/test_base_session.py::test_receive_loop_cancel_notification - AttributeError: 'NoneType' object has no attribute 'message'
===== 1 failed, 10659 passed, 4 skipped, 297 warnings in 311.62s (0:05:11) =====