Add support for LLM testing

[mkutz]

When testing the output of LLMs, outputs are not always deterministic and there can be multiple different, but still approved answers.
Also answers can differ on a low level (e.g. different formulations, filler words , etc.).

Maybe ApproveJ could support multiple approved values/files and allow fuzzy matching in an Approver.