fix: clarify the AI evidence flow

[GwriPennar]

Summary

• replaces the awkward stacked Markdown beneath “Can AI explain the result safely?” with three responsive evidence cards;
• makes the authority chain explicit: exact enumeration, governed Qiskit comparison, then validated GPT‑5.6 explanation;
• preserves every scientific value, evidence contract, explanation safeguard, and API-key boundary.

Root cause

The section used forced Markdown line breaks as a diagram fallback, so its labels and arrows rendered as an unfinished vertical text stack. The replacement uses static, responsive HTML rendered by Streamlit and requires no JavaScript or Mermaid runtime.

Validation

• focused Learning Console tests: 14 passed;
• full pytest: 264 passed;
• Ruff lint: passed;
• Ruff format: passed;
• release verification: passed;
• public-safety scan: passed;
• desktop and 700px visual review: passed;
• no horizontal overflow;
• experiments/** and dependencies unchanged;
• no IBM or OpenAI service contacted.