Description
Current non-overlapping switch logp support is limited to zero-threshold conditions (e.g. switch(x > 0, …, …)), where the branch can often be inferred from the sign of the observed value. Extend this to non-zero thresholds:
y = switch(x > k, f_true(x), f_false(x))
with k a scalar constant (or non-measurable scalar), and constraints ensuring the branch images do not overlap so the observed value implies the originating branch.