Improve logging behavior of DockerOperator

[jscheffl]

We are using DockerOperator for executions outside the cloud... and compared to KubernetesPodOperator I see a lot of problems when logs are emitted. This PR improves the situation:

• If an image is pulled all layer status is posted to logs - mostly this is not of interest, added a cool new log group around
• As logs from container are captured, log chunks are logged to stdout for Airflow. But chunks mostly contain multiple lines and if a lot of logs are emitted then... at point of retrieval the task_log_handler tries to sort messages by timestamp an fails if numeric content is in the docker logs and totally scrambles the result. Logs are with this PR split by line and the python logger prefix is added consistently on each line - like in K8sPodOperator.

[potiuk]

Nice one :)