Revisit how individual cameras are co-added in specsim results

Specsim returns per-camera vectors as well as co-added results. We currently only use the per-camera results in `quickbrick` and `quickgen`, but for SNR-level studies a co-add is still useful.

The coadd is currently throughput weighted:

```
obsflux = sum( thru[c] * resolution.dot(srcflux) ) / sum ( thru[c] )
```

where both sums are over cameras only.

Should we move to ivar weights, or even a full-blown spectro-perfect co-add (a la desi-doc-1056) ?

This is related to #1.

Start by tracing how co-added results are used in quickbrick and quickgen, and review the per-camera vectors and current throughput-weighted calculation described in the issue. Compare the possible ivar-weighted and spectro-perfect approaches, including desi-doc-1056 and related issue #1. Done should mean an agreed co-add method is implemented and its SNR-study behavior is validated.

Written by the indexing model from the issue text.