Put all the tests requiring openai in one place and skip/run them in the pytest command in the workflow

[sonichi]

          In future, we can put all the tests requiring openai in one place and skip/run them in the pytest command in the workflow, instead of skipping each test based on whether openai is installed.

Originally posted by @sonichi in #1052 (comment)

[sonichi]

Also see @maxim-saplin's followup comment

[maxim-saplin]

I would break it down in 2 parts:

• support --skip-openai pytest argument to allow skipping all tests that depend on OpenAI API endpoints
• consolidate all model endpoints used by tests in a single file and make sure tests only use models available through that file

[afourney]

There may be other requirements that incur similar constraints. E.g., some agents might need the Bing API Key. Others might need access to some non-standard dependencies, etc. Whatever we do we should ensure the design is extensible.