Fix pandas-stubs mypy regression in validator output handling

[bgaidioz]

Summary

• replace replace({pd.NaT: None}) with typed-safe pandas null normalization in validator output paths
• keep DataFrame and Series handling behavior the same while avoiding the stricter pandas-stubs replacement-map typing error
• add regression tests for NaT in DataFrame and Series outputs

Why

PR #271 bumps pandas-stubs and makes uv run mypy . fail in CI on src/mxcp/sdk/validator/converters.py because the replace({pd.NaT: None}) pattern is typed more strictly.

Testing

• uv run ruff check .
• uv run black --check --diff .
• uv run mypy .
• env -u OP_SERVICE_ACCOUNT_TOKEN uv run pytest --cov=src/mxcp --cov-report=xml --cov-report=term-missing

Notes

• The full local pytest run only needed OP_SERVICE_ACCOUNT_TOKEN unset because one existing config test assumes that variable is absent.