Make agent simulation scale-generic via one fractal loop + LOD roll-up

To simulate a world from individuals to galaxies without per-entity blowup, do NOT write a separate system per scale. Write ONE scale-generic loop over actors-within-a-container, and connect the levels with two conserving transforms: roll-up (a dense cluster of level-N actors promotes into a single level-N+1 aggregate entity carrying summed power/resources/enemies) and spawn-down (materialize the aggregate back into members near the observer). Separate two orthogonal axes: scale (size: person..galaxy, a column) and resolution/LOD (compute budget: background..foreground). The loop runs on whatever the current LOD exposes -- a few aggregate actors per level, not millions -- so cost stays bounded at any world size (a galaxy war is a handful of nation-blocs; zoom in and it spawns down). Practical wiring: re-query actors PER LEVEL inside the tick so an entity promoted at level N is immediately available at level N+1 in the same tick (a coalition of orgs instantly plays the nation game). Same pattern already used for particle/field power aggregation and population; applying it to the political graph (alliances -> nations -> empires) made empires emergent, not authored.