Add TornadoVM, Ivar Grimstad, Daniel Oh, Susanne Pieterse, and Deep Netts news found via governance review

[jamesward]

Summary

• Add TornadoVM as its own card in Inference & Training. It's already referenced as GPULlama3.java's GPU backend, but its v5.2.0-jdk21 release (already in the News feed) adds AI-specific kernels directly (FP8/BF16 tensor-core matmul, cuBLAS/CUTLASS-compatible BFloat16 arrays, batched FP16 GEMM) — enough to warrant standalone billing per CONTRIBUTING.md's relevance bar.
• Add Ivar Grimstad (Java Champion, Jakarta EE Developer Advocate at the Eclipse Foundation) to People to Follow — he actively speaks on bringing AI to Jakarta EE, complementing the existing Jakarta Agentic AI entry.
• Add a News item for How to Create a Spring Boot Fraud Scoring Service (foojay.io, Jul 31 2026) — a tutorial embedding a pure-Java Deep Netts model directly in a Spring Boot app instead of a separate Python inference server.
• Add Daniel Oh (Java Champion, Senior Principal Developer Advocate at Red Hat) — CNCF Ambassador actively speaking on agentic AI and cloud-native Java, including a JavaOne 2026 session on Java agent patterns and MCP.
• Add Susanne Pieterse (Senior Software Engineer & iSAQB-certified Software Architect at OPEN.nl) — LangChain4j contributor running hands-on workshops teaching Java teams to build reliable AI agents (evaluation, guardrails, observability), speaking at JavaOne 2026.
• Verified every current News item against the "no older than 3 months" rule — all are within the window, so no pruning this round.
• Considered but did not add:
  • "Erupt" (from open PR Add Erupt to the list of resources #60) — legitimate, actively maintained project, but its core identity is a low-code admin/CRUD generator with AI as a secondary module, and the PR overstates its LLM-provider count, so it doesn't fit cleanly alongside the AI-first frameworks in that section.
  • Arconia (Thomas Vitale's Spring Boot devex toolkit) — has an AI-specific Docling document-processing pipeline and pluggable AI-observability conventions, but AI is one feature among several general observability/multitenancy features, not the project's core identity.
  • Vectors (integrallis/vectors, embedded Java vector-search library with a Spring AI VectorStore adapter) — genuinely Java/JVM AI-specific and just announced on Spring's official blog, but only 9 GitHub stars and unproven adoption; too early to clear the site's relevance bar.
  • Deep Netts itself as a standalone Inference & Training entry — only 13 commits and no signs of recent maintenance despite the fresh tutorial coverage; the tutorial is newsworthy, the library isn't yet established enough to list.

Test plan

• SPEC.md updated first, then index.html regenerated to match (per AGENTS.md)
• sitemap.xml <lastmod> bumped
• Verified all new facts (avatars, roles, socials, article contents) via direct fetch, not inferred from URLs
• Confirmed avatar image URLs return 200
• Visual check of the rendered page (no build step / local server available in this session)