Need to clarify provenance of e2mip models

[nbren12]

For some e2mip models, the training details could be clearer. We have some details in the table in the README, but the provenance of checkpoints is not clear.

In particular, it's not clear whether e2mip://fcn and e2mip://dlwp are the same checkpoints as in the papers, or whether they are retrained using modulus. I think e2mip://fcn is from the paper, but dlwp is retrained from scratch using the modulus rewrite. Is that right @ktangsali @NickGeneva?

We should make a docs page with more of these details.