Feature extraction - wrapper around schema.apply_udf

[FelipeAdachi]

Currently, if the user wants to use Langkit for a Feature Extraction scenario, they would neet to run:

import toxicity
from whylogs.experimental.core.udf_schema import udf_schema
import pandas as pd

df = pd.DataFrame({"prompt": ["I love you", "I hate you"]})
schema = udf_schema()

df_enhanced, _ = schema.apply_udfs(df)

Which unnecessarily exposes the user to whylogs' udf_schema and provides a confusing tuple output.

This PR wraps the code above into a langkit.extract function, so it becomes like this:

import langkit
from langkit import toxicity

df = pd.DataFrame({"prompt": ["I love you", "I hate you"]})
enhanced_df = langkit.extract(data=df)

or, for the row case:

import langkit
from langkit import toxicity

row = {"prompt": "I love you", "response": "I hate you"}
enhanced_row = langkit.extract(data=row)

also:

• incidental error handling in hallucination module

[jamie256]

This looks great @FelipeAdachi, thanks!