allclose_raise gives ambiguous error message for keys in different order

[sirmarcel]

When using allclose_raise to compare two identical TensorMaps where the keys are in different order, an error like the following is raised:

metatensor.torch.operations._utils.NotEqualError: the tensor maps have different keys: inputs to 'allclose' should have the same keys names, got '[o3_lambda, o3_sigma, center_type, neighbor_type]' and '[o3_lambda, center_type, neighbor_type, o3_sigma]'

This is confusing, since the keys are the same -- they're just not in the same order. The easiest fix would be to change the message to "should have the same keys names and the same order". The more convenient fix would be to normalise the order of the keys before comparison.