Improve documentation on reshaping SamplerQNN output via interpret Callable

[edoaltamura]

What should we add?

Related to issue #674, this proposes enhancing the documentation of SamplerQNN to explain how to reshape the output via the interpret=None, output_shape=None arguments.

What should be implemented

An additional example titled Reshaping output by the number of classical registers and include in either or both the API documentation and a short tutorial (if both, then reference each other). NB the default reshapes by the total number of registers and the example should explain how to reshape by the number of classical registers only.
The documentation enhancement should appear as a minimum in the API reference of SamplerQNN.

Useful suggestion from xyzdxf

Use 2**circuit.num_clbits as the default output shape, instead of 2**circuit.num_qubits.