allow super_qids to be supplied directly to model instantiation

Right now, the division of qids and subproduct_kwargs is kind of arbitrary. Really, it's just inherited from sofind, where it's nice to use one qid for cases where that fully specifies the map (like an array map) and cases where more info is needed (like a null split map). But that's not fundamental, and ultimately only the super_qids are needed for the noise model. There's only two places where self._qids are needed later, for filenames from sofind...

The motivation for this is it would simplify book-keeping where we want to be able to map from a map (which may be a basic array map, or a null split map) to a noise model: how to call it, and what order its outputs are in. It's a little complicated to do that with this division of qid and subproduct_kwargs, and prone to mistakes.

(Also, this could bubble back up to sofind, where we can just use extra qids and never use subproduct kwargs. I don't think that would require any code changes.)

Start by tracing model instantiation and the later uses of self._qids in the model and sofind-related code. Check how qid and subproduct_kwargs currently reach super_qids, then verify that direct super_qids input supports both array maps and null split maps without breaking filename handling or noise-model bookkeeping.