Add heuristics to ensure unique and correct model names

[SebieF]

It would be great to have a unique identifier for each model that is trained via biotrainer, which should be based on the configuration file, the dataset input and the timestamp of training. It should be added to the result file and as a statement in the beginning of the log file.

Furthermore, it should be ensured that if a model is continued to being trained (as for auto_resume or pretrained_model options), the dataset did not change in the meantime. The dataset input could be used to generate a hash that is checked before the model training is continued. If the hash were included in the model name in some way, this could be even easier to check.