I read the document (finally). It makes two key statements
- Photometric calibration (in relative color) must be know to sub mmag level.
- The average effective filter mean wavelength needs to be known to better than 0.1 nm.
I agree with the conclusion that DESC SN Cosmology requires sub mmag precision for SN Ia lightcurves to reach the full potential of LSST SNeIa. Figuring out how to do that is within DESC's remit and not the Project or Operations. It's also possible, although challenging.
The 0.1 nm specification from the document is a requirement on the knowledge of the mean system bandpass for that filter (i.e., the "effective wavelength of the filter"). Conceptually this is about the issue of when we say that the precision in "g" band is 3~mmag, what did we mean by "g"?
It is not a requirement on determining the edges of the filter response to 0.1 nm on a per-visit basis, which would lie somewhere between "not possible" and "crazy".
While the paper presents its analysis as being optimistic, I think they're missing an opportunity for further self-calibration. In particular, the analysis presented in the paper ignores the fact that SNeIa at a range of redshifts will be being observed simultaneously in each image. So an effective wavelength variation would affect a set of supernovae, and you would know which ones those were.