Added a Probana backend and a real device solver

[toncho11]

• Added a new solver that targets real probabilistic computers.
  • The solver was required because the existing p-kit solvers perform the p-bit updates in software, while with a physical device the updates (in the case of Probana) are performed directly on the hardware.
• Added a backend for the research probabilistic computer Probana. The class implements the USB protocol used by Probana and handles the transfer of J and h.

The Probana backend can be used as an example for the implementaion of other physical probabilistic computers.

[gcattan]

Fair enough. If I understand correctly, it just default to a NumpyBackend when used with the other solvers.
There is one edge case you may need to check, is with the execute function, because Pobrana doesn't have annealing right?

[toncho11]

Thank you for the review.
I think some version of annealing can be implemented in the Probana firmware. I will look into it.

[toncho11]

So the currently implemented version in Probana does not support linear annealing, only constant. This means that the following 2 demos won't work:

• tsp.py
• quantum_tfim.py

But Quantum won't work anyway on Probana, because it requires too many p-bits.

[toncho11]

I think your edge case is now addressed. Probana supports local constant/linear annealing, while unsupported custom annealing_func schedules are rejected explicitly in RealDeviceSolver. The backend also avoids parallel access to the single USB device by handling multiple shots through the physical backend itself.

LGTM