GLM metadata support?

[billbrod]

Should we have better metadata support for the GLM objects? in particular, with predict, it feels weird to pass the glm object a pynapple object with columns and then... drop the columns. would be nice to grab columns if available (which would work with pynapple, pandas, etc.), and similarly for rate.

(columns here are neuron labels)

could also make them optional arguments to the function? but that starts to feel cluttered