Is there a reason to use the Plücker formulation instead of just plain raymaps (o, d) when using them as input embeddings? Seems like you just lose information on where the origin of the ray is. In practice they seem to perform identically.