Is the right answer to switch my build stuff to use the named conda env, or to install packages explicitly into base in order to get the needed notebook extensions and stuff going?

Well, I think it depends on what caused the failure. The console output looks like its not related to the scipipe conda install, so its possible it was unrelated. loadLSST.bash does the right thing auto-magically so the only way things could be going wrong [with the conda env] is if the scipipe conda install is being used without soucing loadLSST.bash.

In fact there were a number of places where I was just using the python binary from the stack without sourcing the whole environment. I fixed that and trimmed places where I was incorrectly/unnecessarily installing Jupyter components into the stack Python (which I don't need to do as Jupyter always runs in the external SCL Python).

I rebuilt and ran through a couple notebooks; it seems to be OK now.