Fix Vertex AI model routing and v1beta1 for preview models

[nyo16]

Summary

• Include model name in build_request_params — The Vertex AI override was missing the "model" key in the params map, causing chat/2 and chat_stream/2 to always fall back to "gemini-2.0-flash" regardless of which model was actually requested
• Use v1beta1 API version for preview/experimental models — Preview models (e.g. gemini-3.1-pro-preview) return 404 on the v1 endpoint and require v1beta1
• Add debug logging for Vertex AI request URLs to aid troubleshooting

Details

The root cause was that build_request_params/3 in Nous.Providers.VertexAI built a params map without a "model" key:

# Before (broken)
params = %{"contents" => contents}

# After (fixed)
params = %{"model" => model.model, "contents" => contents}

The Gemini provider already included this correctly. Without it, chat_stream/2 fell back to the default model name when building the URL.

Test plan

• Verify gemini-2.0-flash (stable) uses v1 endpoint
• Verify gemini-3.1-pro-preview uses v1beta1 endpoint
• Verify the correct model name appears in the URL (not the default)
• Run mix test — existing tests pass
• Doctest examples for endpoint/3 and api_version_for_model/1