Repairing PyTorch Hand objective differentiation

[mikhailnikolaev]

Fixings that are applied due to issue #112

Additional changes:

1. Hand test data flattening for Python
2. Decreasing tolerance for PyTorch to 1e-7 because precision 1e-8 is too small for Hand objective differentiation in complicated case for some sets of input data
3. Some code fixings and refactoring

Graphs

[mikhailnikolaev]

That looks great! The code became simpler 👍

Thanks!