Raco-aliked weights support only 9 layers

[maksymshylo]

Hi. I noticed that raco-aliked weights support only 9 layers.

matcher = LightGlue(features="raco-aliked", n_layers=5).eval().to(device)

RuntimeError: Error(s) in loading state_dict for LightGlue:
	size mismatch for confidence_thresholds: copying a param with shape torch.Size([9]) from checkpoint, the shape in current model is torch.Size([5]).

Would it be possible to get support for < 9 layers in the future?