class AuthBoto performance issue

In the current implementation, AuthBoto recreates the boto3.Session() for every mlflow.log_metric() call. This significantly slows down the program when there are multiple log_metric() calls. I think the AuthBoto class should be able to cache the credentials for a period of time, or reuse the credentials until they're about to timeout. This can significantly improve performance.

Start by locating the AuthBoto class and the boto3.Session() creation used during mlflow.log_metric() calls. Trace how credentials are obtained and expire, then identify existing tests covering AuthBoto or metric logging. Done means repeated logging avoids unnecessary session creation while still refreshing credentials safely before timeout.

Written by the indexing model from the issue text.