Ollama failover UX — show which local model was attempted before switching

[Dhwanil25]

Current behaviour

When Ollama is configured but the requested model isn't loaded, the agent silently fails (returns < 20 chars of output) and switches to the next available provider. The user sees ↻ Provider unavailable but doesn't know which Ollama model was attempted or why it failed.

Expected behaviour

The failover notice should be more descriptive:

↻ Ollama — llama3.2 not found locally (is it pulled?)
  Switched to Anthropic / claude-haiku-4-5

And in the agent card, show a tooltip or expandable section with:

• Which model was attempted
• The error received (e.g. model not found, connection refused)
• A hint: Run: ollama pull llama3.2

Root cause

In multiAgentEngine.ts, streamWithFailover() catches the error and sets the notice string but discards the original error message. The Ollama-specific error (model not found) is lost.

Fix

1. In streamOllama(), throw a descriptive error: throw new Error(Ollama: model '${modelId}' not found — run: ollama pull ${modelId})
2. In streamWithFailover(), include the caught error message in the ↻ notice string
3. Surface the full message in the agent card output