Fixed: add_batch method observer on PyUniformReplayBuffer bug #198

[j0rd1smit]

Fixes issue #198

[nealwu]

Hi, this change breaks a test on https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/examples/v1/train_eval.py, specifically because of metric_utils.py, which uses py_driver: https://github.com/tensorflow/agents/blob/master/tf_agents/eval/metric_utils.py#L72.

The error message is

ValueError: Batch size 1 does not match previously set batch size 2. Make sure your batch size is set correctly in BatchedPyMetric initialization and that the batch size remains constant.

@ebrevdo, do you know what the right fix would be?

[tfboyd]

Closing due to time passed. The mismatch between py and tf version of components is confusing; and is being improved slowly based on customer need.