Original city-unit activation

--unit-activation enables an additional Jev choice after a fresh city review. A running Python session can call session.enable_unit_activation() while the game is paused with clear inputs and no unobserved pending commands. The journal records the capability; only subsequent native checkpoints include the extra owned-unit stack metadata. Historical observations keep their original format.

Jev first chooses Review labor, which closes, observes and reopens the city. That same fresh review permits both labor choices and eligible stationed-unit activation choices. The review itself does not move or activate a unit. If Jev chooses a particular fortified unit, the harness uses the original Units Present icon, observes its Unit Information popup, selects Activate Unit and Close City Screen, verifies the selected radio from original pixels, and confirms it. A native checkpoint then checks the actual unit order and selected-unit identity. Jev makes a separate decision about any subsequent movement or other order.

The icon's unit identity comes from the original linked stack order, not sorted unit IDs or a guessed icon. Only complete reciprocal chains of living owned units on the same tile are exposed. The initial calibration supports the observed single row of one to five units with its Units Present heading, non-veteran fortified ground units, and an observed owned home city. Other layouts and unit variants remain unavailable until separately calibrated. Omission does not mean an action is illegal in Civilization II.

The evidence distinguishes selecting an intent, partial ordinary inputs, completed confirmation, and an observed native result. An uncertain input stops the continuation; it is not automatically replayed. The verifier rechecks the original source images, popup resources, radio pixels, stack identity and native before/after state. The harness does not directly write unit records, orders, game memory or save files; the original game applies the ordinary inputs.