refactor(html): compose through native renderer capabilities

[mosobande]

Result

Replace the fixed Cytoscape graph-viewer direction with a smaller native-renderer composition contract for html-artifact.

• choose the reader job and representation first, then use the selected renderer's native grammar/data/configuration surface;
• reuse the containing project's renderer, package manager, bundler, lockfile, CSP, component shell, and test harness when available;
• otherwise use a task-local build context with exact dependency identities rather than adding a persistent QP renderer application;
• keep supplied source material as data while task-authored renderer configuration/styles/transforms/events remain implementation code;
• preserve a semantic DOM/accessibility fallback when a runtime renderer is unavailable;
• scope embedded component CSS/behavior to the component root instead of taking over the containing document;
• require recurring mechanical evidence plus Kọ Skill's script-boundary gate before introducing a QP wrapper, intermediate schema, or reusable renderer code.

Why the earlier candidate changed

The original #95 proved a working offline Cytoscape viewer, but it also fixed a QP-specific graph envelope, breadth-first layout, styling, controls, and document structure before repeated use showed those were stable mechanical seams. That constrained native Cytoscape capabilities such as classes, compound nodes, task-specific layouts, and reader-specific composition.

This rework keeps the valuable architectural lesson—use mature renderer/build capabilities instead of hand-building everything—but moves information-design choices back to HTML Artifact and the capable agent.

Scope

This PR adds no renderer package, graph schema, browser application, hosted service, runtime dependency, or new public skill. Cytoscape remains a strategic capability anchor, alongside Mermaid, Vega-Lite, D3, Pierre Diffs, Shiki, and future mature tools when they fit the reader job.

The PR is rebased directly on current ori (74a07c55). Exact head: 092a512bf716d84be5afce085964adcb802a1c43.

Proof

• local git diff --check passed;
• Kọ Skill package validation passed on the equivalent pre-merge HTML Artifact surface;
• exact-head GitHub Validate run 33963232951 passed, including package/plugin validation, Skills CLI + Claude compatibility, local-session/uninstall/Akọsílẹ̀ mechanics, portable macOS/Windows mechanics, and the already-shipped browser-control proof.

The prior graph-builder Node/browser evidence is intentionally no longer part of this candidate because that implementation has been removed.

[mosobande]

Separate follow-up noted from local-analysis review: ayewo-igba-ise currently has a user-facing orchestration gap for requested historical-analysis → skill-disposition/remediation flows. I am fixing that independently; #95 remains renderer-composition only.