notebooks/agent_memory: add OAMP support assistant example

[RichmondAlake]

Summary

Adds notebooks/agent_memory/oamp_support_assistant_example.ipynb, an end-to-end customer-support copilot built on Oracle AI Agent Memory (OAMP) and the OpenAI Agents SDK. It follows one damaged-delivery support case from setup through to teardown, and is listed as row 06 in the agent_memory README.

What the notebook demonstrates

• Background extraction — raw messages return promptly while memory extraction runs in an in-process worker
• Hybrid search — exact identifiers (RET-14D) and natural-language queries through one search API
• In-database embeddings via OracleDBEmbedder (DB-resident ALL_MINILM_L12_V2)
• Custom extraction instructions, metadata inheritance + filtering (tenant/case/tags/approval), and $array_contains / nested filters
• Context cards for conversation compaction
• Agent tool use — scoped search, context-card retrieval, typed memory write, and thread-scoped correction as OpenAI function tools
• Update APIs — the agent corrects a communication preference; the app updates thread config and chunked policy
• Exact scope matching (exact_user_match=True) to prevent cross-user leakage
• TTL + deletion for short-lived operational memories
• Chunked semantic indexing with chunk reconciliation to one logical record
• Teardown that removes only this run's RUN_ID-scoped data, keeping the store clean and re-runnable

Notes

• Installs oracleagentmemory from PyPI via pip install --upgrade (no local wheel).
• Credentials are read from environment variables (OPENAI_API_KEY, DB_USER, DB_PASSWORD, DB_CONNECT_STRING); none are stored in the notebook, which ships output-free.
• Validated end-to-end against a local Oracle AI Database 26ai (FREEPDB1) with gpt-5.4: all cells execute cleanly with no errors or warnings, and the teardown leaves the store empty.

🤖 Generated with Claude Code