Add ManufacturingAgentSuite example

[Ryanhuii]

Summary

This PR proposes a minimal ManufacturingAgentSuite example for OPEA GenAIExamples, based on the public WearEdge OPEA Manufacturing reference implementation.

It demonstrates one OPEA-style Gateway/Megaservice pattern across five manufacturing agents:

• maintenance: predictive maintenance / CMMS work-order draft
• iqc: incoming/in-process quality hold / QMS event
• changeover: SKU changeover verification / checklist
• wi: released work-instruction guidance
• hazard: EHS hazard observation

OPEA Alignment

The first PR scope is intentionally small and reviewable:

• route registry for five manufacturing modes
• deterministic no-model demo path for CI and quick review
• Qdrant vector DB profile
• optional official OPEA TEI embedding profile using TEI_EMBEDDING_ENDPOINT and EMBEDDING_COMPONENT_NAME=OPEA_TEI_EMBEDDING
• /v1/agents, /v1/agents/{mode}/demo, /v1/agents/{mode}/infer, /v1/scorecard
• guardrails that block unsupported claims such as final root cause, restart permission, maintenance release, quality release, area safe, and safety clearance

Related Discussion And Evidence

• RFC issue: RFC: ManufacturingAgentSuite blueprint for route-isolated industrial action-card agents #2461
• TEI update comment: RFC: ManufacturingAgentSuite blueprint for route-isolated industrial action-card agents #2461 (comment)
• Reference implementation: https://github.com/davidmillerak2026-sys/wearedge-opea-manufacturing
• GCP C3 official TEI E2E evidence: https://github.com/davidmillerak2026-sys/wearedge-opea-manufacturing/blob/main/docs/gcp-c3-opea-tei-profile-e2e-report.md

Validation

Prepared and smoke-tested in the reference repository before opening this PR:

• Python syntax parse for the standalone service
• local HTTP smoke for /healthz, five demos, five infers, and /v1/scorecard
• Docker Compose base config
• Docker Compose official OPEA TEI config

The broader reference implementation has also been fresh-clone validated on Google Cloud C3 c3-standard-4 single-node 4-vCPU / 16-GiB-RAM / no-GPU profile, inside the challenge resource constraint of single node, <=64GB RAM, and 4-core CPU.

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues found.

Scanned Files

None

[Ryanhuii]

Maintainer review request update:

The broader WearEdge OPEA Manufacturing reference package has been further
consolidated after this PR was opened. The latest evidence summary is tracked
here:

davidmillerak2026-sys/wearedge-opea-manufacturing#2 (comment)

Highlights:

• official OPEA TEI local and GCP C3 evidence
• five-route Manufacturing RAG
• optional reranker profile
• strict local Ollama gemma4:31b benchmark with fallback_count=0
• strict DeepSeek deepseek-chat OpenAI-compatible benchmark with fallback_count=0
• official OPEA GenAIEval chatqnafixed local benchmark summary with zero failures
• optional Kubernetes manifest

Boundary:

Please review this PR as a minimal CI-friendly GenAIExamples addition. It does
not claim official OPEA publication before merge, and it intentionally excludes
GraphRAG and fine-tuning/SFT/DPO/PPO for this release path.