Verify OpenAI examples run with the GPT-5.4 generation

[ccozad]

Part of the post-update verification pass. Model strings were standardized on the June 2026 OpenAI generation (see `docs/analysis/03-model-updates.md`). Confirm each example still runs end-to-end with the new models and that output looks correct.

Now using `gpt-5.4-nano` (cheap RAG / completion demos):

• `lang-chain/RAG/rag_pipeline.py`
• `lang-chain/RAG-serve/custom_pipelines.py`
• `lang-chain/docker-serve/custom_pipelines.py`
• `lang-chain/Azure/rag_pipeline.py` (also exercises Azure AI Search)
• `llama-index/llm/app.py`
• `llama-index/citation-query-engine/citation_query_engine.py`
• `lang-chain/fast-api/` (enum + default `gpt-5.4-nano`)

Now using `gpt-5.4` (agentic / tool-using):

• `browser-use/find_car_prices.py`
• `browser-use/find_flights.py` (planner now `gpt-5.4-mini`, was `o3-mini`)

How to verify: set `OPENAI_API_KEY`, set up each example's venv per `docs/setup/python-venv.md`, run, and confirm sensible output. Watch for: model-id rejected by the API, response-format changes, and (browser-use) tool-calling/planning still working.