[voice] Mission-context situational narration (CASEVAC, HLZ, golden hour, etc.)

[jdev-02]

Idea from Kyle (P3) during the entities ontology PR.

Today (#21 baseline): TTS speaks the route's physical rationale.

'Routed to Lobos Creek, distance 2.1 kilometers, ETA 38 minutes on foot covered.'

Goal: speak the mission-aware rationale.

'Based on your need for CASEVAC, I'm routing you to an open field suitable for an HLZ that can allow for the casualty to be extracted within the golden hour.'

This requires:

1. Mission context input — not just NL prompt + lat/lon, but also the role of the request (CASEVAC, evac, hasty defense, medevac, etc.). Probably an enum on PlanRequest or inferred from the LLM's RouteQuery (mission_type already supports it).
2. Domain vocabulary — CASEVAC, HLZ (helicopter landing zone), golden hour (60-min trauma window), bound, knee, wadi, etc. Lives in ontology/vocabulary.yml.
3. Narrative templates — per mission_type, a template that strings together the route facts + the mission rationale. Currently only one generic template in agent/orchestrator.py _build_rationale.
4. Awareness of selected entities — pulling from entities.yml: 'open field' = landuse_open + sufficient size; 'covered' = landuse_forest; etc.

Owner: Jon (P1).
Priority: P2, stretch. Hero demo doesn't need it. Post-Phase-3 if time.
Source: Kyle, Sat ~1620 in chat.