fix(bigquery): dbapi socket leak

[chalmerlowe]

Problem

The system test test_dbapi_connection_does_not_leak_sockets fails intermittently because network connections remain in the ESTABLISHED state even after connection.close() is called. This is likely due to the underlying gRPC (remote procedure call) transport not being immediately closed, or delays in Garbage Collection (cleaning up unused memory).

Solution

This Pull Request updates Connection.close() to explicitly close the underlying gRPC channel (transport._grpc_channel) if it exists on the BigQuery Storage client. This ensures that the connection is closed immediately and resources are released deterministically.

Notes to Reviewers

• This change explicitly targets the underlying gRPC channel to bypass potential delays caused by transport interceptors or non-deterministic garbage collection.
• It addresses intermittent test failures caused by lingering ESTABLISHED sockets.
• Also includes several linting revisions (sorry).

Sample debug info from traceback:

E --- Socket Leak Debug Info ---
E Start Count: 6
E End Count: 7
E Current Connections:
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=48550), Raddr: addr(ip='74.125.195.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=53976), Raddr: addr(ip='173.194.43.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=54818), Raddr: addr(ip='74.125.199.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=43762), Raddr: addr(ip='142.251.188.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=53946), Raddr: addr(ip='173.194.43.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=54792), Raddr: addr(ip='74.125.199.95', port=443)
E Status: ESTABLISHED, Laddr: addr(ip='10.138.0.14', port=48524), Raddr: addr(ip='74.125.195.95', port=443)

[gemini-code-assist]

Code Review

This pull request updates the Connection.close() method to explicitly close the underlying gRPC channel of the BigQuery Storage client's transport, preventing potential socket leaks. It also refactors unit tests to mock the public transport property instead of the private _transport attribute. The review feedback suggests using mock.patch.object with self.addCleanup to prevent test pollution when mocking the transport property, and using getattr instead of hasattr to safely check for _grpc_channel and avoid potential AttributeErrors if the channel is None.

[daniel-sanche]

LGTM