Revisit: Suggest interactions in the statistical model disambiguation phase + interface

[emjun]

See previous discussion here.

If analysts do not provide interacts, should rTisane suggest interactions in the statistical model disambiguation phase? This could be helpful for promoting deeper reflection, but comes with challenges in (i) generating appropriate (all?) possible interaction effects (or maybe: is there a smarter subset of interactions to suggest?) and (ii) surfacing/showing/guiding end-users through the possible effects in the interface.