Inspect if we need types_mapper when converting from arrow to pandas in stream()

[hagenw]

In #542 we removed the types_mapper argument when converting a arrow table to a pandas dataframe. The type_mapper was simply copied from the audformat implementation, where it seems to lead to an arrow when we remove it, compare audeering/audformat#508

[hagenw]

The types_mapper in audformat is needed for the case when we load from a CSV file with pyarrow. Here, in audb.stream() we always use pd.read_csv() in which case types_mapper is not needed. We also test for it, as the current test runs for paruet and csv files.