fix: raise ValueError when no classification is found

[grgkovac]

Summary

This PR addresses #1638 by improving the error handling when a classification metric cannot find a specified (or default) classification ID in the DataDefinition.

The Problem

Currently, if a user defines a BinaryClassification with a custom name, but forgets to specify that classification_name in their Metric, the system returns None. This results in a cryptic KeyError: None deep in the library.

See issue #1638

Code example:

import pandas as pd
from evidently import Dataset, DataDefinition, Report, BinaryClassification
from evidently.metrics import Accuracy

data = pd.DataFrame({"id_col": [1, 2, 3], "actual": [0, 1, 1], "predicted": [1, 0, 1]})
classification_ = [BinaryClassification(target="actual", prediction_labels="predicted", name="cls")]
data_definition = DataDefinition(id_column="id_col", classification=classification_)
dataset = Dataset.from_pandas(data, data_definition=data_definition)
report = Report(metrics=[Accuracy(target="actual", prediction="predicted", name="acc")]) # <- missing: classification_name="cls"
report.run(dataset)

The Fix

I modified get_classification in datasets.py to raise a descriptive ValueError instead of returning None. This follows the "fail fast" principle and provides the user with actionable feedback.

New Error Message:
ValueError: No classification with id 'default' found.

I can potentially add something like this as well?
If you are using custom names, ensure you pass 'classification_name' to your Metric (e.g., Accuracy(..., classification_name='your_name')).

---

Fixes #1638