Add PINN mode

[SuperSashka]

It is good to have possibility to add something like data=data and term in a loss function to be able to train model in a PINN mode.

[SuperSashka]

I would like to make a term inside the loss function that based on (optionally) its own grid and its own data make a competition with a differential equation with its own weight. Consequently, we need separate computation procedure which may be included to loss function, i.e. L=lam_data * L_data+lam_op * L_op+lam_bc * L_bc

[SuperSashka]

At the time being we have separate 'data' BC type #36 that has its own weight and may be included into adaptive lambda procedure.

However, I still think that it may be improved further, so it is not closed atm.