feat: add per-frame 2D visualization (static demo) (#11)

src/tracking/evaluation/interactive.py
  - _ggiw_to_ellipse(): GGIW extent → {ext_a, ext_b, ext_theta, x, y, vx, vy}
  - run_once(): add record_frames flag; captures GT/obs/est snapshots per frame
  - run_simulation_json(): include frame_data (sensor_origin + frames[]) from MC run 0

scripts/generate_demo.py
  - Runs run_simulation_json for selected/all scenarios
  - Writes docs/data/demo_<scenario>.json

docs/demo.html
  - Canvas 2D renderer: GT rectangles (green), LiDAR points (yellow),
    GGIW estimate ellipses (red), velocity arrows, sensor origin marker
  - Frame slider + Play/Pause/Step/Reset controls
  - Auto-fit coordinate transform with grid and axis labels
  - Missed-detection badge for miss frames
  - Per-scenario summary metrics strip
  - Loads all scenarios from docs/data/ via fetch (falls back gracefully)

tests/tracking/test_interactive.py (+10 tests)
  - frame_data present and correctly nested
  - frame index sequential, GT fields valid, obs pairs finite
  - est ellipse axes positive (guards eigenvalue sign bugs)
  - missed frames have empty obs + missed=True flag
  - record_frames=False → snapshots is None
  - record_frames=True → snapshots length matches n_frames

Usage (local preview):
  uv run python scripts/generate_demo.py --all
  cd docs && python -m http.server 8080
  # open http://localhost:8080/demo.html

https://claude.ai/code/session_013sQREGvTmjgFT8cyRNmPUL

Co-authored-by: Claude <noreply@anthropic.com>

Author: YoshiRi