Allow user to pass continuous observation to the loss function

When RBM is used in an ML pipelines like DVAE it is necessary to update encoder parameters with respect to energy / free energy of the RBM over the output of encoders which are continuous and carry gradients during backpropagation.
The energy / free energy should be calculated with parameters of RBM (h, J) fixed (detached).
specifically the `quasi_objective` function in here [https://github.com/dwavesystems/dwave-pytorch-plugin/blob/main/dwave/plugins/torch/boltzmann_machine.py#L323](url) should allow an optional parameter (continuous observation) and return the related energy / free energy.