Add Plotly result-analysis notebook and logical-change narrative notebook

[Copilot]

Adds two focused notebooks requested in the issue: one for graph-based result interpretation using Plotly, and one documenting what changed and why at a logical/design level.

• Problem framing

  • Introduce review-ready artifacts that make performance outcomes easy to inspect visually.
  • Capture architectural and workflow changes in a concise, decision-oriented narrative (not code walkthroughs).

• Notebook deliverables

  • notebooks/results_analysis_plotly.ipynb
    • Loads baseline/candidate benchmark reports.
    • Produces Plotly views for:
      • side-by-side metric comparison
      • accuracy/latency Pareto tradeoff
      • prioritized bottleneck comparison
  • notebooks/logical_changes_explained.ipynb
    • Summarizes the key logical shifts: acceptance criteria, reproducible benchmarking, profiling-first optimization, data pipeline redesign, training/runtime strategy controls, model variant strategy, and CI guardrails.
    • Includes a lightweight acceptance-summary cell reading benchmark artifacts.

• Repo/documentation updates

  • Added plotly to requirements.txt.
  • Updated README.md with notebook locations and minimal commands to generate required benchmark artifacts consumed by the notebooks.

# Results notebook artifact inputs
BASELINE = ROOT / "benchmark_runs" / "baseline" / "benchmark_report.json"
CANDIDATE = ROOT / "benchmark_runs" / "candidate" / "benchmark_report.json"

fig = px.scatter(
    df_trials,
    x="inference_latency_ms",
    y="mean_iou",
    color="run",
    size="throughput_images_per_sec",
    title="Accuracy-Speed Pareto View (Lower latency, higher IoU)",
)