Maintain global column types in returned dataframe when date_range or row_range filter out all rows

[alexowens90]

When a column in a returned dataframe has no values in it (e.g. if dynamic schema is enabled but this column was not present in the the date range requested) we return a column with dtype float64 full of NaNs in Pandas 1.X, and with dtype object filled with Nones in Pandas 2.X, matching Pandas behaviour for empty columns in both cases.

In the special case where there are no rows at all, we should maintain the dtype of the column as it is for that symbol-version pair in storage.