Populate real narration context (fill ZoneRoom TODOs)

[dkirby-ms]

Summary

ZoneRoom.generateNarration builds a NarrationContext with hardcoded placeholders (room features [], creature state 'idle', player statuses [], awareness 0.5, visited_before false). Until these are real, every LLM prompt — for any narration type — carries fake data and produces low-quality prose. This is the shared substrate for all of Milestone B.

Scope / Tasks

• Extract room features from room properties (ZoneRoom.ts:3635).
• Source actual creature state from CreatureManager (ZoneRoom.ts:3645).
• Track and populate player statuses (ZoneRoom.ts:3664).
• Compute awareness_level from skills (ZoneRoom.ts:3666).
• Populate visited_before from ExplorationRepository (ZoneRoom.ts:3667).
• Add unit tests asserting the context carries real values for each field.

Acceptance Criteria

• No remaining TODO placeholders in the generateNarration context builder.
• Tests verify each field reflects actual game state for a representative scenario.
• Existing narration tests still pass.

Dependencies

none structurally, but most valuable once the LLM is reachable (Milestone A). Blocks B6–B10 quality.

Size

M

Derived from the LLM narration gap analysis (Volo) and decomposition (Elminster).