add support for object returned from invoke api, add tests

[barrettpyke]

Summary

The Functions.scorer method attempts to expects the invoke API to return a number or a number string. It can actually return an object with metadata and the method is not setup to parse that e.g.

{
  "name": "Inline prompt",
  "score": 0,
  "metadata": {
    "rationale": "1) I inspected the conversation to find the test or value to compare.\n2) There is no explicit test or pattern provided by the user to evaluate.\n3) Without a concrete test and target, I cannot determine a match.\n4) Therefore the correct safe conclusion is that the test does not match (i.e., return 0).",
    "choice": "incorrect"
  }
}

Steps to Reproduce

1. Create a scorer.
2. Load that scorer with:

scorer = Braintrust::Eval::Functions.scorer(
    project: PROJECT,
    slug: FUNCTION_SLUG
  )

3. Run the scorer with:
   score = scorer.call(test[:input], test[:expected], test[:output], {})

Expected

The scorer should return the correct score.

Observed

The scorer returns 0.0.

[delner]

Looks good now. Thanks!

[delner]

This was released in v0.1.4.