[Bug]: Almost all tests fail to check for empty config_lists

[afourney]

Describe the bug

As seen in #1076 many tests fail to check for valid config_lists, and silently default to GPT-4. Even if we are ok with it using GPT-4, this situation silent failure is suboptimal.

Impacted are tests for RetrieveChat, CompressibleAgent, and GPTAssistant.

Steps to reproduce

No response

Expected Behavior

No response

Screenshots and logs

No response

Additional Information

No response

[afourney]

#1082 should now help with this as well.