examples: add Ollama persistent-memory recipes

[ferhimedamine]

Problem

Ollama's /api/chat and /api/generate are stateless — every call starts fresh, so Ollama apps have no cross-session memory. Users kept asking how to add persistent recall (e.g. ollama/ollama#16987). The SDK already ships ChatMemorySession, but we had no worked Ollama example to point people at.

Design

Two complementary, runnable examples under examples/:

• ollama_memory_chat.py — embeds recall→chat→store directly around ollama.chat using ChatMemorySession. Recalls across the agent's full memory (not just the current session), stores each turn as episodic memory, and weights a durable user preference at high importance so it reliably resurfaces on later, semantically related turns.
• ollama_memory_proxy.py — a transparent FastAPI middleware. Point any existing Ollama client at :8080 instead of :11434 and /api/chat transparently gains memory with no application changes. This implements the recipe requested in Community recipe: Persistent memory for Ollama apps via Dakera middleware ollama/ollama#16987, with the grounding corrected: Dakera listens on port 3000 (not 3300), self-hosting uses the dakera-deploy compose (the server needs its object store), and recalled context is injected as a system message rather than a non-existent top-level system field.

Both use only the documented public SDK surface (DakeraClient / AsyncDakeraClient, ChatMemorySession / AsyncChatMemorySession, store/recall/close).

Testing

• ruff check . — clean (repo config).
• python -m py_compile — clean.
• Logic verified under mocks: chat() injects the system context, forwards to Ollama, and stores the user+assistant turns in order; _last_user_message() handles present/absent/empty user turns.
• Not added to the live-server examples CI job: unlike the other examples it requires a running Ollama daemon with a pulled model, which the CI service container can't provide. Each example guards its optional third-party imports (ollama, fastapi/httpx) with an install hint.

Checklist

• Focused, additive change (examples only — no package/API changes)
• Follows existing examples/ style (memory.py, hybrid_search.py)
• Lint clean; README Examples list updated
• Links relevant issue (Community recipe: Persistent memory for Ollama apps via Dakera middleware ollama/ollama#16987)

[ferhimedamine]

🤖 [Agent: Core Engine] Post-merge verification (review requested via Paperclip 87ab1b51; PR was merged before it landed): all SDK references verified against source — ChatMemorySession/AsyncChatMemorySession exports and signatures (create, store(role, content, importance=0.6, tags=), recall(query, top_k=5), close(summary=)), RecalledMemory.content, and the port-3000 default (matches server default config). Examples-only, CI green, public-repo hygiene clean. ✅ No follow-up needed; example link is live on main.