BigQuery: Allow choice of compression when loading from dataframe

[plamut]

Closes #7701.

This PR adds an optional parameter to Client.load_table_from_dataframe() method that allows selecting the compression method if directly serializing a dataframe to a parquet file.

How to test

The issue description should be self-explanatory, but mind one of the comments:

Also I'm actually a bit wary of exposing compression, since I'd like to keep the option open to change the file format we serialize to. Parquet happens to be the best match to BigQuery's data types right now, but it'd be good to keep the option open to move to something else in the future.

I opted for the name parquet_compression instead of compression to indicate that the parameter is specific to parquet serialization, and not applicable in all use cases of the method.

[tswast]

We do need to modify _pandas_helpers.dataframe_to_parquet as well. See:

google-cloud-python/bigquery/google/cloud/bigquery/_pandas_helpers.py

Line 230 in 1b3e822

pyarrow.parquet.write_table(arrow_table, filepath)

The underlying pyarrow.parquet.write_table function also takes a compression argument.

Long-term, I expect the _pandas_helpers.dataframe_to_parquet function to get used more often than the dataframe.to_parquet method. We'll want to start fetching the table schema if not provided and use that for pandas to BigQuery type conversions #8142.

[plamut]

There is a mismatch between the two methods, the pyarrow's accepts a richer range of compression methods.

Having either two different compression parameters, or a single parameter that accepts different values depending on the context, could be confusing to the end users, thus I will only allow the compression methods supported by both.

However, if there are uses cases that would need specific support for LZO, LZ4, and ZSTD as well, please let me know. Probably there aren't, because to date the compression method has not been exposed anyway?

It's good that we are marking the parameter as beta, as I can see how this can change in the future. 👍

Update:
Changed my mind after realizing that we probably should document the underlying serialization methods and link to their original docs. Since we are already exposing that detail, it makes less sense to try hiding compression options behind the lowest common denominator.

[tswast]

Thanks.

I agree that it's confusing to expose this level of internal details, but I think your docstring description is clear.