agents-workshop: fix agent_setup failing on deprecated claude-3-7-sonnet PPT

[Incognico-o]

Summary

• databricks-claude-3-7-sonnet was removed from pay-per-token serving, causing the agents-workshop/agent_setup/agent.py notebook to fail at deploy time with:

  NotFoundError: 404 - {'error_code': 'NOT_FOUND', 'message': 'Pay-per-token for this model is disabled. Please see Databricks documentation for supported Foundation Models.'}
  

• Swapped the active LLM_ENDPOINT_NAME to databricks-gpt-oss-120b — the same model used in Part 2 (02_agent_eval/agent.py), so the full workshop now runs on a single consistent model.
• Left databricks-claude-sonnet-4-6 commented out as a quick Claude fallback alongside the existing databricks-meta-llama-3-3-70b-instruct fallback.

Why gpt-oss-120b

• Confirmed live on PPT in e2-demo-field-eng (and across the FE workspaces lab attendees use).
• Matches the model already shipping in 02_agent_eval/agent.py — students see one model end-to-end.
• Supports tool calling, which the LangGraph bind_tools(tools) path requires.

Test plan

• Run agents-workshop/agent_setup/agent_setup.py end-to-end and confirm mlflow.pyfunc.log_model(...) succeeds.
• Confirm agents.deploy(...) produces a working agents_agents_lab-product-product_agent endpoint.
• Sanity-query the deployed endpoint with the existing input example ("What color options are available for the Aria Modern Bookshelf?") and confirm it tool-calls into the vector search retriever.

This pull request and its description were written by Isaac.