Introduce a parameter `input_gradients` to NeuralNetwork

[adekusar-drl]

What is the expected behavior?

As a feedback from #44, we should add a parameter input_gradients to NeuralNetwork to expose it explicitly to users. By default it should be turned off. Also, we should not turn on input gradients explicitly in TorchConnector every time, turning on/off gradients is up a user via this new parameter.

[jonvet]

mind if I take a look at this one?

to confirm:

• NeuralNetwork and all classes that inherit from it should have a new parameter input_gradients
• should the current getter/setter in NeuralNetwork be deprecated?

[adekusar-drl]

I don't mind, any help is appreciated.

Things to do:

• Add input_gradient as a constructor parameter to NeuralNetwork with a default value of False. So, yes, this is the answer on the first question.
• Verify there are getter/setter in place (as I recall there's just a getter). So, no, current getter/setter are not deprecated.
• Remove a line in TorchConnector where input_gradients is set to True since this should be done explicitly by a user. Probably this should be mentioned in the documentation(docstrings).
• Add a line to set input_gradients to True in the torch connector tests.
• May be more unit tests are required, can't say now.
• Add a reno file with the changes.

Also take a look at https://github.com/Qiskit/qiskit-machine-learning/blob/main/CONTRIBUTING.md Probably that's it. Let me know if you have questions.

[jonvet]

cool thanks, i also have permission from my employer now, so I'll get started