This hampers using envs that prefer `conda-forge` over the `defaults` channel even from within a JupyterLab that is running in a `defaults`-only `base` env.