Extends base fern client with eval utilities

[peadaroh]

This PR extends the auto-generated Humanloop client with additional utilities for running an Evaluation where the user manages their full application runtime.

Related fern docs for extending client with custom code: Augment with custom code — Fern

Specifically we add a method humanloop.evaluations.run_local(...) that takes details of a file destination on Humanloop, a user defined dataset, a user defined function to evaluator, a name for their Evaluation and details of their Evaluators.

This is invoked like so:

from my_code import ask_question, load_dataset
checks = hl.evaluations.run(
    file={
        "path": "evals_demo/answer-flow",
        "type": "flow",
        "function": ask_question,
        "version": {"attributes": {"description": "Simple RAG, OpenAI + Chroma", "prompt": PROMPT}},
    },
    name="Staging CI",
    dataset={"path": "evals_demo/medqa-test", "datapoints": load_dataset()},
    evaluators=[
        {"path": "evals_demo/exact_match"},
        {"path": "evals_demo/levenshtein"},
        {"path": "evals_demo/reasoning", "threshold": 0.5},
    ],
    workers=4
)

With resulting output:
The resulting CLI output is

If the file or dataset does not yet exist on Humanloop, they will be created automatically, otherwise they will be updated appropriately.

If the Evaluation exists for the file (based on the name provided) then a new run is added, otherwise a new Evaluation is automatically created.

Evaluators referenced by path (or id) must already exist. However, we've also provisionally added an affordance for local (or external) Evaluators:

from my_code import ask_question, load_dataset
from my_evaluators import levenshtein_distance_optimized

checks = hl.evaluations.run(
    file={
        "path": "evals_demo/answer-flow",
        "type": "flow",
        "function": ask_question,
        "version": {"attributes": {"description": "Simple RAG, OpenAI + Chroma", "prompt": PROMPT}},
    },
    name="Staging CI",
    dataset={"path": "evals_demo/medqa-test", "datapoints": load_dataset()},
    evaluators=[
        {"path": "evals_demo/exact_match"},
        {"path": "evals_demo/levenshtein"},
        {"path": "evals_demo/reasoning", "threshold": 0.5},
        {
            "path": "evals_demo/levenshtein-optimized",
            "function": levenshtein_distance_optimized,
            "args_type": "target_required",
            "return_type": "number",
        }
    ],
    workers=4
)