Change `test_pypdf_loader_w_images()` to free multimodal model

[erwallace]

test_pypdf_loader_w_images() currently uses openai/gpt-4o-mini. Should switch to a free model to make sure it doesn't cost to run tests every time.