check that scripts with last statement as SELECT can be downloaded with `to_dataframe`

[tswast]

I'm uncertain how the destination table is populated for script jobs. It's possible we'll need some logic to extract the destination table from a child job if we want to support the BigQuery Storage API in this case.