We avoid using the common names fixed effects and random effects, which are not only misleading but also defined differently across the various fields in which these models are applied. We instead favor the more accurate (albeit more verbose) intercepts/coefficients that are common across groups and intercepts/coefficients that vary by group. See more at
Estimating Generalized (Non-)Linear Models with Group-Specific Terms with rstanarm
http://andrewgelman.com/2010/12/17/so-called_fixed
Why I don’t use the term “fixed and random effects” | Statistical Modeling, Causal Inference, and Social Science
I’m not fluent in the lmer formula syntax, but if I read it correctly, in your second model you have group specific distance and context, so the model still has those available. Now is seems that groups are so different that common term is not useful and the groups have so much data that the group specific terms are very accurate. p_loo being around 72 also supports tha hypothesis. See this http://www.stat.columbia.edu/~gelman/research/published/final_sub.pdf for related discussion,
Aki