Feature/265 train request checking with validator

• added a TrainingArgs class in common/training_args.py
• replaced the training_args with this class on researcher side
• added a serializer to pass the TrainingArgs to the node side

WARNING:

• on node side, training_args is still a dict
• main reason: the researcher can add an extra_scheme (or even modify the current default scheme) on researcher side (e.g. to implement testing/poc features), which node will not be aware of and as a consequence will not have the same scheme to check the TrainingArgs against..