feat(workflow): the solver review rejects a step too simple to justify itself

[EdmondDantes]

What

The solverReview critic — which gates a generated solver — now rejects a workflow that contains a step too simple to justify its own existence. The generator prompt already says "make meaty steps, no ceremony", but that is a rule in a prompt the model can ignore; the critic is the gate that actually catches it.

Why this shape (not a turn count)

It is an evaluation, not a measurement. An earlier idea was to count a step's model round-trips at runtime and flag anything under ~15 turns — dropped: the threshold is arbitrary, a live step legitimately varies, and a runtime check fires after the work is done, too late to fix the decomposition. Whether a workflow is well-split is knowable from the structure, at generation time, before anything runs — so the reviewer judges it by reading the step breakdown and asking, per step: "if it does this little, is it a step at all, or ceremony?"

The rule added to the rubric

A step whose prompt carves out only a sliver — a lookup, a restatement, one trivial action a neighbour would absorb without noticing — should be folded into that neighbour; a step must justify its own fresh context. Kept explicitly distinct from step count: having few steps is not the fault (a lone meaty step is ideal); a thin step is.

Test theSolverReviewRejectsAStepTooSimpleToJustifyItself pins the criterion in the reviewer's prompt.