Replace hardcoded default model with dynamic Ollama detection

[bussyjd]

Summary

• Remove hardcoded gpt-oss:120b-cloud default model (does not exist) from OpenClaw overlay and llmspy config
• Add listOllamaModels() to query the host's Ollama for actually pulled models during SetupDefault()
• Populate the OpenClaw overlay with real available models; set agentModel to the first one
• When Ollama has no models pulled, deploy with empty model list and guide users to obol model setup or ollama pull
• Update llmspy llms.json default model from gpt-oss:120b-cloud to empty string

Test plan

• go test ./... passes
• With Ollama running and models pulled: obol stack up deploys OpenClaw with detected models in overlay
• With Ollama running but no models: deploys with empty model list, prints guidance
• Without Ollama: skips OpenClaw setup, prints guidance to run obol model setup