What changed

• Added an explicit window < 2 guard in _rolling_std to return an all-NaN result immediately.
• Expanded the _rolling_std docstring to clarify the sample-std (ddof=1) behavior for undersized windows.
• Tightened test_window_1_returns_nan to assert the undefined window=1 case returns all NaN values.

Why

• Sample standard deviation with ddof=1 is undefined for window sizes below 2.
• The previous implementation still called np.std(..., ddof=1) for window=1, which emitted runtime warnings during tests.
• Returning early keeps behavior mathematically correct and removes avoidable warning noise.

Testing

• ✅ scripts/clone_and_test.sh massive-com/mcp_massive (583 passed)