use temp directory to save models trained through API

[wochinge]

Proposed changes:

• models trained with POST /model/train now get saved in a temporary directory by
  default (instead of the models directory)

Status (please check what you already did):

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

[wochinge]

@ricwo
Being able to specify the out with the request seems pretty dangerous to me. You could use that to overwrite any files on the server you want to, couldn't you?

output_path=rjs.get("out", tempfile.gettempdir())

I think we should remove the option to do so.

[ricwo]

Looks great

[wochinge]

@ricwo https://docs.python.org/3/library/tempfile.html#tempfile.gettempdir stays the same over multiple runs. It could only lead to problems, when running multiple rasa instances on one machine, but I think we can neglect that for now (the worst case is that a training is done although it could have been skipped). What do you mean?

[ricwo]

yes that sounds good, let's use that - alternatively we could introduce a boolean save_to_rasa_default_model_dir which defaults to True. that way the default behaviour would be unchanged. when firing training requests from rasa X we would set it to False. what do you think?

[wochinge]

Yep, good idea! Will do the change tomorrow to rasa and X

[ricwo]

Looks great 🚀 Thanks for making this backwards-compatible