better hydra integration

[simplymathematics]

• [ ] move optimizer/loss function to Model dataclass
• [ ] use post-init instead of call functions
• [ ] pre-declare properties (i.e. data_file, training_time) that are defined during the post-init
• [ ] minimize code differences between exp/opt
• [ ] read dvc.params in on ever layer but optimise
• [ ] parse dvc params using hydra to simplify the params.yaml output
• [ ] better hydra integration (remove remaining name/params/kwargs)
• [ ] remove redundancies in call functions