rm option loses some logs

Hi everyone,

Problem:

I've found a problem. When we are using docker run with option auto_remove. Somethimes we can skip some logs.

Example:

For example lets run some python code with a bug:
Let's use docker cli:

1. without rm option

aleksandermangin@ubuntu ~/dev/src/docker-py (master) $ docker run python python -c "xxx"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
NameError: name 'xxx' is not defined

2. with rm option

aleksandermangin@ubuntu ~/dev/src/docker-py (master) $ docker run python python -c "xxx"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
NameError: name 'xxx' is not defined

So looks like great.

Let's do the same staff with docker-py:

1. without rm option

In [1]: import docker                                                                                                                        
In [3]: c = docker.from_env()                                                                                                                                                                               
In [6]: c.containers.run('python', 'python -c "xxx"', stderr=True, stdout=True)                                                                                                                             

ContainerError: Command 'python -c "xxx"' in image 'python' returned non-zero exit status 1: b'Traceback (most recent call last):\n  File "<string>", line 1, in <module>\nNameError: name \'xxx\' is not defined\n'

1. with rm option
   In [1]: import docker
   In [2]: c = docker.from_env()
   In [3]: c.containers.run('python', 'python -c "xxx"', stderr=True, stdout=True, auto_remove=True)
   ContainerError: Command 'python -c "xxx"' in image 'python' returned non-zero exit status 1

So we lost a part of message with rm option!

Where is the problem?

https://github.com/docker/docker-py/blob/master/docker/models/containers.py#L826

Here we did two bad things:

1. We changed the rules of logging. I think It's so surprising behaviour
2. We changed the rules after running docker container. We are losing here a lot of messages.

How to fix it?

My pr will be soon

Start in docker/models/containers.py around line 826, where the issue reports different error output when auto_remove is enabled. Reproduce the two containers.run examples with and without auto_remove and compare the complete error message; done means the auto-remove case preserves the logged failure details.