feat(gs-ai): add AI telemetry health explainer

[root3315]

What

Adds an optional AI assistant to the ground station that turns decoded subsystem telemetry into a plain-language operator status report.

Why

Operators currently read raw decoded fields (battery SOC, CPU temp, pointing error…) and judge health manually. This module produces an at-a-glance natural-language assessment and is a clean hook for an LLM-backed narrative — without making the ground station depend on a network service.

Design

• Rule-based engine is the default backend. Deterministic thresholds for EPS (battery SOC, net power), OBC (CPU temp, error count), ADCS (pointing error), GNSS (fix) roll up to a nominal → advisory → warning → critical severity ladder. Runs fully offline, so tests are deterministic and CI needs no secrets.
• Optional Google Gemini backend (use_gemini=True + api_key) produces richer narrative text. The optional google-generativeai import is lazy and isolated.
• Graceful degradation: any Gemini failure (missing library, no network, bad key) falls back to the rule-based text — operations never break.

Changes

File
ground-station/utils/ai_telemetry.py	New module (rule engine + Gemini backend), mypy-strict clean
ground-station/tests/test_ai_telemetry.py	10 unit tests (rule paths + fallback + exception hierarchy)
ground-station/pages/12_ai_assistant.py	Demo Streamlit page with sample scenarios
ground-station/tests/test_pages_smoke.py	Page count 11 → 12
ground-station/requirements.txt	Optional google-generativeai documented (commented)

Gates

• pytest — 118 passed, 1 skipped (10 new)
• black --line-length 79 — clean
• mypy --strict utils/ai_telemetry.py — clean
• No existing module touched; new code is additive

---

Generated by Claude Code