Refine data pipeline modularization report

[piotrnowakowski]

a full overview of what we have and should do for easier picking apart of the code implemented by agents from this repo into another projects

[piotrnowakowski]

I think i introduced it in previous PR but ignore the diff as the document was not fully ready before it had much to long context that was not usefull for example

[mircorudolph]

I agree with the report.

I think the most important steps are:

• Add missing benchmarks where operational quality gaps are largest
  -> this aligns strongly with writing detailed artifacts. Each step should create clearly named artifacts with inputs, outputs, and relevant intermediate values. Those should always be written as opposed to on-demand benchmarks
• Add explicit result models for stage outputs where tuple or loose-dict
  contracts are doing too much work. -> makes a lot of sense to have clear hand-off contracts.