Improve API for third party compilers.

[erick-xanadu]

In order to implement CUDA, I had to override some functionality in Catalyst.

        with Patcher(
            (catalyst.pennylane_extensions.QFunc, "extract_backend_info", cudaq_backend_info),
            (qml.QNode, "__call__", QFunc.__call__),
        ):

I think third party compilers might have to do something similar if they follow a similar approach to CUDA-quantum's implementation.

[dime10]

I'm not sure this actually concerns Catalyst, but is something to look at for PennyLane's compiler API.