[Bug] QuantumNaturalGradient not compatible with trainable feature maps

[DanieleCucurachi]

When optimizing a QuantumModel using QuantumNaturalGradient, an error occurs if a feature map with trainable parameters is utilized. The following error message is encountered:

Intel MKL ERROR: Parameter 6 was incorrect on entry to DGELSD. Working... 

...

File "/home/...python3.10//site-packages/qadence_libs/qinfo_tools/qng.py", line 158, in step qng_spsa(vparams_values, vparams_keys, grad_vec, lr, circuit, self.state, epsilon, beta) 
File "/home/.../python3.10/site-packages/qadence_libs/qinfo_tools/qng.py", line 231, in qng_spsa transf_grad = torch.linalg.lstsq( 
RuntimeError: false INTERNAL ASSERT FAILED at "../aten/src/ATen/native/BatchLinearAlgebra.cpp":1537, please report a bug to PyTorch. 
torch.linalg.lstsq: Argument 6 has illegal value. Most certainly there is a bug in the implementation calling the backend library.

This issue appears to stem from incompatible dimensions or invalid input to the least squares solver. It seems that the current Quantum Natural Gradient (QNG) implementation does not properly account for the additional trainable parameters of the feature map.