feature: Add Python-native DataPipe interface for fluent data preprocessing

[mathysgrapotte]

Is your feature request related to a problem? Please describe.

Python users currently need to rely on YAML/config files for defining data preprocessing pipelines, which can be cumbersome for interactive experimentation and native Python workflows.

Describe the solution you'd like

A fluent Python interface (DataPipe) that enables chaining of data processing operations (split-transform-encode) while maintaining compatibility with existing config-based workflows. The interface should provide a clear pipeline construction pattern similar to Nextflow processes but in native Python.

[mathysgrapotte]

very vague suggestion :

pipe = (DataPipe(raw_df, loader)
        .split(RandomSplitter, ratios=[0.8, 0.2])
        .transform(AddNoise, columns=['ecg'], std=0.1)
        .encode(LabelEncoder, column='diagnosis')
        .build())

dataset = HandlerTorch(pipe.get_tensors()).to_dataset()

but we have to see about this once refactoring is done