Migrate healthcare-assistant demo app from galileo to splunk-ao SDK

[etserend]

Summary

• Migrates all three `healthcare-assistant` workshop variations (`2-app-with-instrumentation`, `3-app-with-experiments`, `4-app-with-controls`) from the `galileo` SDK to `splunk-ao`
• Updates credentials (`GALILEO_` → `SPLUNK_AO_`), callbacks (`GalileoAsyncCallback` → `SplunkAOAsyncCallback`), logger (`GalileoLogger` → `SplunkAOLogger`), context manager (`galileo_context` → `splunk_ao_context`), and k8s secret references (`galileo-secret` → `splunk-ao-secrets`)
• Updates Agent Control SDK deps (`agent-control-sdk[galileo]` → `agent-control-sdk[splunk-ao]`) and API key header (`Galileo-API-Key` → `Splunk-AO-Key`) in 4-app
• Updates all corresponding workshop content pages (en + ja) with new package names, credential names, image references, and console UI steps
• Adds `OPENAI_EMBEDDING_BASE_URL` support in `rag.py` and `k8s.yaml` so Azure deployments that use separate chat/embedding endpoints work correctly
• Patches both sync and async OpenAI clients in `app.py` to inject `api-version` query param required by Azure APIM
• Wraps hallucination demo retriever+llm spans in a workflow span so the Splunk AO console shows a single named root trace instead of two orphan spans (2-app and 3-app)

Validation — Phase 1 (Galileo SDK baseline) ✅

Ran full test session against Galileo staging REST via `_validate_galileo.py`:

• Lisinopril RAG query → 24 spans, full retriever tree with real context docs
• Patient P001 SQL query → 10 spans, SQL generation LLM call captured
• Hallucination demo → 2 flat root spans (retriever + llm), as expected for Galileo SDK

Validation — Phase 2 (splunk-ao SDK → Galileo staging REST) ✅

Ran full test session via Streamlit UI at localhost:8540:

• All three interactions confirmed working end-to-end
• Span structure matches Phase 1: no spans dropped, retriever output populated
• Naming differences are systematic OTel verb prefixes (`invoke_workflow`, `execute_tool`, `retrieval`) — cosmetic only
• Hallucination demo now shows a single `invoke_workflow Hallucination Demo` root — improvement over Phase 1

Test plan

• `2-app-with-instrumentation` starts and traces appear in Splunk AO console (validated Phase 2)
• RAG query returns answers from knowledge base with retriever spans populated
• SQL query returns patient data with SQL-generation LLM span captured
• Hallucination demo logs successfully with workflow root span
• `3-app-with-experiments` dataset creation and experiment run complete with `SplunkAOEvaluators` metrics
• `4-app-with-controls` Agent Control initializes on startup, blocking/steering controls work
• Workshop content pages render correctly (no broken image refs, correct code snippets)

Closes HYBIM-977