Transform decorator

[MatsMoll]

Made it easier to add different types of udfs

@feature_view(source=InMemorySource.empty())
class CombinedData:
    query = String()
    contains_mr = query.contains('mr')

    @transform_polars(using_features=[query], return_type=Bool())
    def contains_something(self, df: pl.LazyFrame, return_value: str) -> pl.LazyFrame:
        return df.with_columns((pl.col('query').str.len_chars() > 5).alias(return_value))

    @transform_pandas(using_features=[query], return_type=String())
    def append_someting(self, df: pd.DataFrame) -> pd.Series:
        return df['query'] + ' something'

    @transform_row(using_features=[query], return_type=String())
    def using_row(self, row: dict) -> str:
        return row['query'] + ' something'

    not_contains = contains_something.not_equals(True)

Also made it easier to switch between a model's output source and when to use the exposed model with

# Reading from the output source
using_source = store.model(MyModel).predict_over(...)

# Sending the input features to a model
use_model_store = store.without_model_cache()
preds_sent_to_model = use_model_store.model(MyModel).predict_over(...)