[Parity] Commute stress is tower-wide, not per-Sim: a lone far tenant in an idle tower under-stresses

Surfaced by the game/design party while ratifying the routing parity change (#503/#509).

The 1994 original computes tenant evaluation as Quality = 300 - (total stress / tenants), where each Sim accumulates stress from ELEVATOR WAIT TIME across its whole commute (a 6-transfer trip waits at 6 landings, carrying 6 waits of stress). See the FAQs/wiki.

Our engine models the wait cost as a tower-wide rolling frustration (crowd.frustration, the fraction of travellers past STRESS_WAIT) plus per-floor congestion feeding gripes/move-outs. That is a faithful AGGREGATE proxy and moves the right direction, but it is not the original's per-Sim accumulated stress.

The gap only bites an edge case: a lone far tenant whose specific multi-transfer commute is slow while the rest of the tower's elevators sit idle. Congestion reads ~0 everywhere, tower frustration ~0, so that tenant is under-stressed relative to the original.

This is PRE-EXISTING (it existed at MAX_RIDES=2 too; uncapping reachability merely exposes it, because far tenants can now move in). Not a blocker for the parity fix. Fidelity refinement: accumulate a per-origin-unit commute stress from the per-person wait we already compute, and let it feed that unit's satisfaction. Both modes benefit; Modern's legible version is the #502 track.

Start by tracing the per-person wait calculation and how crowd.frustration, STRESS_WAIT, and per-floor congestion feed unit satisfaction. Compare that path with the original total-stress-per-tenant behavior and the #502 Modern track. Done means both modes account for accumulated multi-transfer wait stress, including the lone far-tenant case.