`_cook_pp`: trailing `NaN` in `pw_sem` after `ffill()` causes spurious "pp_df has missing values" warnings

[samuelwnaylor]

In _cook_pp, after invalid bins are nulled out, pw_sem_{pre/post} is filled with ffill():

pp_df[pw_sem_col] = pp_df[pw_sem_col].ffill()

ffill() only propagates values forward (low to high wind speed). If the last N bins near cutout wind speed are all invalid (insufficient data), they have no subsequent valid value to propagate from, so they remain NaN. This causes the "pp_df has missing values" warning to fire on every turbine/reference pair, including every bootstrap sample, even when there is no meaningful data quality issue.

Potential fix

Add bfill() after ffill() to fill any remaining trailing NaNs with the nearest valid value below cutout:

# before
pp_df[pw_sem_col] = pp_df[pw_sem_col].ffill()

# after
pp_df[pw_sem_col] = pp_df[pw_sem_col].ffill().bfill()