feat: make semantic action graph the primary incident view

[ling101w]

Summary

• replace the dashboard-first homepage with a real-data semantic action graph incident console
• map security overview, audit records, and enforcement settings into nodes, edges, request context, and a selection-specific inspector
• add primary attack path and full graph modes, draggable highlighted nodes, edge selection, and synchronized timeline/event-result navigation
• promote the incident conclusion and Prompt Injection evidence while reserving red for confirmed attack behavior and green for effective enforcement
• update visible product branding to 玄鉴 and include responsive visual QA evidence

Backend compatibility

• keeps the existing detection and enforcement backend unchanged
• consumes the existing /api/security/overview, /api/records, and /api/settings/enforcement endpoints
• fetched and compared against origin/main; there are no upstream-only commits or merge conflicts, and this branch is the frontend source of truth

Verification

• 26 targeted dashboard and security-screen tests passed
• npm run build
• npm run typecheck
• npm run lint
• git diff --check
• design-qa.md: final result: passed

CI note

• GitHub reports this PR as mergeable with no file conflicts.
• The openclaw-plugin workflow is also failing on the current main base SHA. The PR reproduces the same existing Semantic Judge failures (semantic-adversarial timeout and two semantic-runtime-boundaries assertions); this frontend-only change does not touch those backend paths.