Cannot load dates outside of pandas nanosecond timestamp range with `load_parquet`

[tswast]

googleapis/python-db-dtypes-pandas#45

Possibly we need a fallback for out-of-range values that converts to object dtype of datetime.date instead? Or a pandas series containing pyarrow date scalars?

[tswast]

Catching the out-of-bounds and fallback to object seems to make the most sense. Possibly we don't even need to manually convert all the rows and just leave the data as-is for these? It'd mean folks couldn't write strings to DATE when in that far-out range, but maybe that's OK?