Capability-Gated Planning: Cost-to-Goal Discovery and the Limits of Myopic Experiment Selection