Add headless renderer validation pipeline

[bobthearsonist]

Goal

Turn the PR #7 side/headless renderer repro into a repeatable validation pipeline so renderer regressions are caught before we rely on manual Obsidian screenshots.

Why

During PR #7 recovery, Cytoscape runtime data had all 12 fixture nodes/11 edges, but Obsidian visibly painted only 7 nodes when using label-sized nodes. The working fix was validated by a mix of static renderer contracts, layout metrics, and Obsidian screenshots. We need that side-render loop formalized so basic MVP parity stays protected.

Scope

Build a no-surprises renderer validation script that can run locally and in CI/package validation:

• Render a frozen Profisee/MVP sidecar fixture outside Obsidian using the plugin renderer/style path or a faithful Cytoscape harness.
• Assert exact fixture structure: 12 nodes, 11 edges, expected IDs, valid endpoints, non-empty edge labels.
• Assert renderer contract: flat sidecar nodes before edges, no story-card compound parents, no render-time deterministic layout, visible labels/arrowheads, delayed fit strategy.
• Assert visual/runtime output: all nodes exist, all edges exist, node positions are preserved, no compound parents are present, labels are non-empty.
• Produce a stable artifact under a repo-owned validation directory for debugging failures.
• Add a package script such as npm run validate:renderer:headless and include it in the appropriate validation path once stable.

Suggested implementation

• Reuse the current scripts/renderer-contract.mjs fixture/contract checks as the source of static assertions.
• Promote the previous headless Cytoscape repro into a maintained script under plugins/obsidian-plugin/scripts/.
• Prefer existing dependencies and system Chrome/Chromium first; only add Playwright if the no-new-deps approach proves too brittle.
• Keep generated screenshots/HTML out of source control and clean them between runs.
• Fail with a clear message if a local browser is unavailable.

Acceptance criteria

• npm run validate:renderer or a sibling script catches the known regressions:
  • synthetic story-card parent nodes
  • render-time layout/backfill
  • hidden weak/long edge labels
  • label-sized-node paint regression where runtime has 12 nodes but output visibly drops nodes
• The script documents where its transient artifacts are written.
• The validation can be run without opening Obsidian.
• Existing npm run validate, npm run build, and npm run lint remain green.