use tf without gpu support by default

[wochinge]

Proposed changes:

• TensorFlow 1.15 for some reason includes the binaries for cpu and gpu which makes our docker images huge. Switching to the cpu only package.
• Rasa part of https://github.com/RasaHQ/rasa-x/issues/1643
• more information here https://www.tensorflow.org/install/pip

Status (please check what you already did):

• added some tests for the functionality
• updated the documentation
• updated the changelog (please check changelog for instructions)
• reformat files using black (please check Readme for instructions)

[dakshvar22]

So, if I understand correctly, installing tensorflow==1.15.0 brings binaries for both cpu and gpu but pinning to tensorflow-cpu does not?
Do we also support a GPU based docker image?
If yes, we should pin to tensorflow-gpu there.

[wochinge]

@dakshvar22 So tensorflow included only the cpu packages until 1.15. tensorflow==1.15 included both gpu and cpu packages. For tensorflow 2 it's again the old pattern: tensorflow==2.0 is only cpu, while tensorflow-gpu==2.0 is the gpu package. So by using tensorflow-cpu==1.15 we just go back to the way we have been releasing things before.

[wochinge]

@dakshvar22 Can I merge this?

[dakshvar22]

@wochinge Sounds good!