Model effects as emergent per-actor processes, not global knobs

When an adoption/technology effect is modeled as a global multiplier keyed on a diffusion count, it is a hack that describes the outcome instead of the process. The fix: make the effect a consequence of which capability each actor actually operates (owned, paid-for, fuelled), and let cost/behavior read per-actor. Concretely, delete the global scaling factor and gate graph edges by whether the actor can run that mediums transport. Bonus: if the global was always the identity value in existing worlds, deleting it keeps the test suite byte-for-byte unchanged, and the new per-actor logic only diverges for the novel case.