Pass min_periods to rolling() in WindowFeatures

[VenishPaneliya]

What this fixes

WindowFeatures documents min_periods as a pandas passthrough:

min_periods: int, default None. Minimum number of observations in the window required to have a value; otherwise, the result is np.nan. See parameter min_periods in pandas rolling() documentation for more details.

It is accepted and stored as self.min_periods, but transform() never reads it. Both branches build the window without it:

X[self.variables_].rolling(window=win)          # list-of-windows branch
X[self.variables_].rolling(window=self.window)  # single-window branch

So rolling() keeps its default — a full window is required — and the leading rows stay NaN no matter what the user passes.

ExpandingWindowFeatures already does forward it (.expanding(min_periods=self.min_periods)), so the two transformers disagreed on a parameter they document in the same words.

Evidence

A 6-row frame, window=3, default functions="mean", periods=1:

	result
min_periods=None	[nan, nan, nan, 2.0, 3.0, 4.0]
min_periods=1 (before)	[nan, nan, nan, 2.0, 3.0, 4.0] — identical, flag inert
min_periods=1 (after)	[nan, 1.0, 1.5, 2.0, 3.0, 4.0]
X["x"].rolling(3, min_periods=1).mean().shift(1)	[nan, 1.0, 1.5, 2.0, 3.0, 4.0]

After the change the output matches pandas exactly, for a single window and for a list of windows.

Compatibility

The default is unchanged. min_periods=None is precisely what rolling() was already assuming, so anyone who never set the parameter sees identical output. Only users who explicitly passed it see a change — and it is the documented behaviour they asked for.

Testing

Added test_min_periods_is_used and test_min_periods_is_used_with_multiple_windows, comparing against pandas rolling(...) directly and asserting the default still requires a full window.

• without the fix: both fail
• with the fix: both pass
• tests/test_time_series: 121 passed
• flake8 and black clean on both files