Estimating causal moderation effects with randomized treatments and non‐randomized moderators