Add scale argument to `LieGroup.rand` and `LieGroup.randn`

[luisenp]

It would be useful for rand and randn to receive a scale argument to control the magnitude of the sampled groups. From offline discussion, one possibility is to make scale: Optional[Union[float, torch.Tensor]]=None , where if a float is passed, all sampled data is scaled by this value, and if a tensor is passed, it will have the translation then rotation convention, with a single value for all rotations, and dim (2D or 3D) values for the translation component.