Pad Each Store's Own Calendar, Not the Global One

A sales table only has a row for a day something sold. A model needs a row for every day, including the zeros, and where each store's calendar starts is a fact about that store, not a fact about the dataset.

Write fill_missing_days(df). Given a DataFrame of daily sales with store_id, date (datetime64) and sales (float), return a DataFrame with one row for every calendar day from that store's own first recorded date to its own last recorded date, inclusive — filling any day with no recorded sale as 0.0.

A store that opened later than another, or closed earlier, must not gain fabricated rows outside its own observed range. Return columns store_id, date, sales, sorted by store_id then date.