[BUG] Empty columns are converted to `RangeIndex` in `df.set_index(...)` function

Reproducer:
```python
import cudf

gdf = cudf.DataFrame()
gdf['date'] = cudf.Series([], dtype='datetime64[ms]')
broken_gdf = gdf.set_index('date')

print(broken_gdf.index)
print(broken_gdf.index.dtype)
```