Avoid eager NumPy checks for prepared parameters

[adsharma]

Summary

• Avoid pybind NumPy array checks for non-NumPy prepared statement parameters
• Avoid importing NumPy in the C-API fallback unless the value type is from NumPy
• Keep ndarray parameter support intact for users that do have NumPy installed

Root Cause

The ndarray support added NumPy-specific isinstance checks to the prepared parameter path. In pybind, checking against py::array can initialize NumPy even for ordinary Python parameters.

Validation

• uvx ruff check src_py/_lbug_capi.py
• PYTHONPATH=./build .venv/bin/pytest test/test_prepared_statement.py test/test_issue.py::test_issue_483_numpy_ndarray_parameter -q
• Fresh pybind smoke test with numpy imports blocked via sys.meta_path while executing a scalar prepared statement