BigQuery: `load_table_from_dataframe` automatically generate schema for known dtypes

[tswast]

Is your feature request related to a problem? Please describe.

I want to upload a DataFrame without specifying a schema. My DataFrame has well defined dtypes for all columns, so it should be safe to serialize the DataFrame without worrying about cases such as ARRAY / STRUCT columns.

With #9042, I may get a warning in this case, though.

Describe the solution you'd like

I'd like BigQuery to automatically generate a schema based on the dtypes in the DataFrame. Ideally, this generated schema could be merged with the partial schema at #8140.

If the number of columns for the generated schema + partial schema matches that of the DataFrame (meaning, we handled all the object columns with an explicit schema), there should be no warning message printed.

Describe alternatives you've considered

• Status quo: let pandas/pyarrow/fastparquet figure out what to do.
• Introspect the actual objects in the object column to guess the correct type. (Slow!)

/CC @plamut