Lognormal constraint poor convergence

I think the difference between the two cases, is the clip introduces a discontinuity / flat region in the gradient but does not avoid the sampler for seeing it, whereas your original uniform completely avoids the impossible region.