`compact_incomplete` fails if one of the dataframes is of empty-type

[alexowens90]

v1 API, behaviour almost certainly the same with finalize_staged_data in the v2 API. Minimal repro:

lib.write(sym, pd.DataFrame({"col": ["hello"]}, index=[pd.Timestamp("2000-01-01")]), parallel=True)
lib.write(sym, pd.DataFrame({"col": [None]}, index=[pd.Timestamp("2000-01-02")]), parallel=True)
lib.compact_incomplete(sym, False, False)

gives InternalException: E_ASSERTION_FAILURE Allocate data called with zero size