feat: LLM router with local (Ollama) and cloud (OpenAI) backends

- lyra.config.load() reads env into a frozen Config dataclass
- lyra.llm.complete(messages, backend) routes to Ollama /api/chat or
  OpenAI chat completions
- lyra.llm.embed(texts) calls OpenAI embeddings
- .env.example switched from Anthropic to OpenAI to match available key

.env.example

@@ -2,9 +2,10 @@
 LOCAL_BASE_URL=http://localhost:11434
 LOCAL_MODEL=qwen2.5:7b-instruct
 
-# Cloud backend (Anthropic Claude) — used for harder reasoning.
-ANTHROPIC_API_KEY=
-CLOUD_MODEL=claude-sonnet-4-6
+# Cloud backend (OpenAI) — used for harder reasoning and embeddings.
+OPENAI_API_KEY=
+CLOUD_MODEL=gpt-4o-mini
+EMBED_MODEL=text-embedding-3-small
 
 # Where Lyra stores her memory.
 LYRA_DB_PATH=data/lyra.db