feat: use BigQuery Storage client by default

[plamut]

Closes #86.
Closes #84.

No issue yet, just a POC preview.

The PR is now ready to be reviewed. It uses the BQ Storage API by default to fetch query results, and removes the "beta" label from that feature.

It's probably easier to review this PR commit by commit.

Key points:

• If BQ Storage client cannot be used (e.g. missing dependencies) to fetch query results, the library falls back to the REST API.
• The BQ storage API v1 stable is now used, replacing v1beta1 client (however, there is a feature request to support both BQ Storage client versions simultaneously).
• The Python DB API now more closely follows the specs. Closing a connection instance makes it unusable from that point on, and also closes every Cursor instance created by it.
• Closing a DB API Connection now also closes the BigQuery / BQ Storage clients the latter has created, fixing the sockets leak discovered.

[plamut]

I did some timings to compare fetching data with and without the BQ Storage client, results below. Spoiler: When using the BQ Storage API, performance improvements are enormous.

Query

SELECT * FROM `project.dataset.table`
LIMIT 200000

Source tables

• A synthetic table with two FLOAT columns containing 10M rows of random numbers
• bigquery-public-data-cms_medicare.inpatient_charges_2011 - multiple numeric and string columns, contains ~163k rows.

Timings

	REST API (tabledata.list)	BQ Storage API
Synthetic table	10 - 11.5 s	4.5 - 5 s
Inpatient charges table	23 - 25s	10.5 - 12.5 s

[shollyman]

Semantic satiation seeing all the v1/v1beta references. Sorry it took me so long to get to this. Couple minor nits, with a couple more interesting bits (avro vs arrow, small result set, versioning for storage).

[shollyman]

Thanks for the slogging through this.

[plamut]

Just a thought - since this is a pretty significant change, how about releasing all the other smaller changes and fixes that have accumulated since the last release separately, and release this one separately for an easier rollback, should that be necessary?

[shollyman]

Isolating this change to its own release seems prudent.

[plamut]

Putting this on hold until the next release is made, we will ship it separately.

[plamut]

With the new release (1.25.0) now out, we can now merge this and release in the near future (possibly with miscellaneous related cleanup fixes).