Required regression-based loss functions for kernel training.

[SaashaJoshi]

What should we add?

I have found that a TrainableFidelityQuantumKernel is incompatible with regression algorithms, like QSVR. The reason seems to be that TrainableFidelityQuantumKernel utilizes QuantumKernelTrainer class that only supports SVCLoss (this class uses SVC internally) for optimizing the quantum kernel parameters.

Is there a specific reason why Qiskit Machine Learning only provides implementation of SVCLoss and no other alternative, regression-based loss functions? Is this a known limitation?

TO-ADD: Regression-based KernelLoss functions.