Include the master in num_worker_replicas in the contrib.run_config.

This behaviour would be in line with core.run_config.  `num_worker_replicas` is a natural input for constructing a SyncReplicasOptimizer instance with `replicas_to_aggregate=num_worker_replicas` and currently that requries adding +1 in the Cloud ML setting.  Please see here: https://github.com/tensorflow/models/pull/2343.

There is a bit of conflict in terms, since we imply a distinction between a "master" and a "worker" elsewhere, but we wish to include master in workers when querying for num_worker_replicas.  I attempted to improve this by adding a comment to the description of the `master` constructor argument.

I browsed through internal and external github codebases and I didn't notice this breaking anyone, but we are about to find out :-)

PiperOrigin-RevId: 168702095