google-auth: Thread safety race condition during 401 client certificate rotation in transports

When handling `401 Unauthorized` responses, the HTTP transport interceptors in `google.auth.transport.requests` and `google.auth.transport.urllib3` attempt to auto-rotate expired mTLS client certificates by calling `configure_mtls_channel()`.

In multi-threaded environments, if multiple threads share an authorized session and hit a 401 concurrently, they will trigger the certificate rotation at the same time. Because there is no synchronization lock, this leads to race conditions. Multiple threads will redundantly read the updated certificates and concurrently mutate the underlying session's adapter mappings (`requests.Session.adapters`) or pool managers, which is not thread-safe and can cause runtime exceptions.

We should introduce a lock (e.g., using `threading.Lock`) around the `configure_mtls_channel` reconfiguration block inside the transport handlers to ensure that certificate rotation is performed serially.

Start in google.auth.transport.requests and google.auth.transport.urllib3, focusing on the HTTP transport handlers that call configure_mtls_channel after a 401 response. Trace the shared-session and pool-manager reconfiguration paths, then verify that concurrent certificate rotations are serialized and no adapter or pool-manager mutation occurs simultaneously.

索引モデルが issue の本文から書いたものです。