[Parity]: [P2] Per-room EVAL parity and inspector reasoning text

Imported from the engineering backlog (_bmad-output/implementation-artifacts/backlog.md, story per-room-eval-parity).
Priority P2 (do-first order) | Severity: med | Status: idea | Epic: SimTower parity | Type: feature-request | Recorded: 2026-07-08

Divergence from the 1994 original

Owner observations 2026-07-08 (real vs imported, same rooms). Two related gaps in per-unit evaluation: (A) the EVAL VALUE is computed differently from 1994 (owner saw the same room score differently in each; ties to the satisfaction-map difference where our imported tower reads uniformly high/green while the real one has spread). (B) MISSING FEATURE: the 1994 inspector shows textual reasoning for a unit's eval, e.g. a Condo popup reading "Single Room neighbor is noisy" (a full sentence naming the dominant dissatisfier). Add per-unit eval REASONS surfaced in our inspector (noisy neighbor, poor elevator access/long wait, too few nearby amenities, dirty hotel, etc.), derived from the same factors the satisfaction model already weighs. Note our imported satisfaction starts high because we don't import live satisfaction/people (see population-census-parity, tdt-import-population-seed) and re-simulate from content, so compare eval parity on a SETTLED tower, not right after import. Reconcile our satisfaction factors + weights against the 1994 model before claiming a calc bug. Player-facing: reasons make the "why is this unhappy" legible (pairs with our existing satisfaction heatmap, PR #118).

Start with the per-room-eval-parity story in _bmad-output/implementation-artifacts/backlog.md, then read the existing satisfaction model and inspector alongside the satisfaction heatmap from PR #118. Compare a settled tower with the 1994 behavior, including evaluation values and dominant dissatisfiers. Done means per-unit evaluation parity is reconciled and the inspector surfaces readable reasons such as noisy neighbors, elevator access, amenities, or hotel cleanliness.