- Add an enum
- Rm support for multi-criteria in single evaluator

Use `EvalConfig` n the `run_on_dataset` function, plus:
- Add proactive validation for compatibility with the evaluators ( e.g.,
check can be converted to prompt or messages for LLM or check example
input keys against the chain)
- Improve error messaging for the dataset <-> model you're testing
- Integration tests for the combinations of dataset formats and llm,
chat models, and chains


Deltas off [#7508](#7508)
(split the criteria evalutaor into the reference free and labeled
classes) which builds off
[#7388](#7388) which migrates
from langchainplus_sdk to langsmith package

<details> <summary>Dataset/ model setup</summary><pre><code>
#  """Evaluation chain for a single QA evaluator."""
from uuid import uuid4
import pandas as pd
from langchain.client.runner_utils import run_on_dataset

from langsmith import Client

client = Client()

dataset_name = f"Testing - {str(uuid4())[-8:]}"
df = pd.DataFrame(
    {
        "some_input": [
            "What's the capital of California?",
            "What's the capital of Nevada?",
            "What's the capital of Oregon?",
            "What's the capital of Washington?",
        ],
"some_output": ["Sacramento", "Carson City", "Salem", "Olympia"],
    }
)
ds = client.upload_dataframe(
df, dataset_name, input_keys=["some_input"], output_keys=["some_output"]
)


from langchain.chat_models import ChatOpenAI
from langchain.chains import LLMChain


def chain_constructor() -> None:
    """Evaluate a chain on a dataset."""
    llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
    chain = LLMChain.from_string(llm, "What's the capital of {input}?")
    return chain
</code></pre></details>



**Relevant snippet**
```
from langchain.evaluation.run_evaluators.config import (
    RunEvalConfig,
)

evaluation_config = RunEvalConfig(
    evaluator_configs=[
        RunEvalConfig.Criteria(criteria="helpfulness"),
        RunEvalConfig.Criteria(
            criteria={"my-criterion": "Is the answer fewer than 10 words?"}
        ),
        "qa", # Or could do RunEvalConfig.ContextQA(), etc.
        "context_qa",
       "embedding_distance",       
    ]
)


run_on_dataset(
    dataset_name,
    llm_or_chain_factory=chain_constructor,
    run_evaluator_config=evaluation_config,
)

``