apps: add supplychain-demand-planning-agent

[RichmondAlake]

Summary

Adds a new app entry — supplychain-demand-planning-agent — under apps/.

A multi-agent demand-planning assistant on Oracle AI Database. A LangGraph supervisor decomposes planner requests and delegates to two specialists (demand_analyst and policy_agent). Every memory layer, every retrieval primitive, and every LLM call lands in a single Oracle database:

• OracleVS — vector knowledge (12 demand reports + 1 buy-volume policy)
• AsyncOracleStore — long-term cross-thread memory (planner preferences)
• AsyncOracleSaver — per-thread agent checkpoints
• OracleSemanticCache — LLM-response cache
• OracleChatMessageHistory — chat-session transcripts

Embeddings are produced in-database via an ONNX model (ALL_MINILM_L12_V2, 384 dims) loaded into Oracle with DBMS_VECTOR.LOAD_ONNX_MODEL — no external embedding API in the chain.

Bundle includes:

• FastAPI + WebSocket backend that streams supervisor.astream_events("v2") as typed UI events
• React + Vite + Tailwind frontend with chat pane, per-agent context window, live data explorer, animated architecture explorer, and light/dark theming
• Setup scripts (bootstrap.py, onnx_setup.py, seed_supplychain.py)
• Student / complete workshop notebook pair (9 TODOs across 12 parts) under workshop/
• Per-part docs under docs/

Also adds one row to the Apps table in the root README.md.

Test plan

• docker compose up -d oracle-free brings Oracle Free up healthy
• python app/scripts/bootstrap.py creates the AGENT user and the vector memory pool
• python app/scripts/onnx_setup.py loads ALL_MINILM_L12_V2 into Oracle and a smoke-test query returns a 384-dim vector
• python app/scripts/seed_supplychain.py writes 12 demand reports + 1 policy memo into OracleVS and 2 user preferences into AsyncOracleStore
• python -m uvicorn app.backend.main:app --host 0.0.0.0 --port 8000 boots; /api/health returns 200 with {"llm_provider": "oci", "llm_model": "xai.grok-4-1-fast-reasoning"}
• cd app/frontend && npm install && npm run dev -- --port 3000 binds 3000; UI loads with the chat pane, data explorer (all 8 seeded tables visible), and architecture explorer
• All three starter prompts produce a Priya-preference-aware buy recommendation in the chat
• workshop/notebook_complete.ipynb executes end-to-end with 9/9 TODO checkpoints passing