Handle determinism in the pipeline

[ravinkohli]

This issue is being moved from the old repository. Look here for more details.
Description:
we need to add support for setting seeds for various random number generations. Could be handled once in the api for the whole project or individually for pipeline components.

@LMZimmer suggested:
For this let's:

1. create a set_random_state in the BasePipeline
2. call it in the init after [this](https://github.com/automl/Auto-PyTorch/blob/5c6ce0bbf030fb4ec81396e1cb67edb3e194d383/autoPyTorch/pipeline/base_pipeline.py#L104)

add support for this in the smbo part and pass the state to the pipeline

[renesass]

Moreover, as far as I can see, random state is not used for the train data loader (in base_data_loader), which uses shuffling. Correct me if I'm wrong.

[ravinkohli]

Moreover, as far as I can see, random state is not used for the train data loader (in base_data_loader), which uses shuffling. Correct me if I'm wrong.

Yes that is true. Thanks for pointing this out