Design and implement scaffolding for pipeline evaluation

[shadeMe]

Implement scaffolding code that:

• Accepts...
  • The evaluated pipeline, i.e., the pipeline whose output is to be evaluated.
  • A set of inputs for the above pipeline.
  • The evaluation pipeline, i.e., the one with the evaluation components/metrics.
  • A set of additional inputs for the evaluation pipeline, e.g: labels, etc.
• Runs...
  • The evaluated pipeline with the above inputs.
    • Optionally allows overriding parameters of specific components in said pipeline.
  • The evaluation pipeline with the outputs of the above pipeline.
• Returns...
  • The results of the evaluation pipeline.

The scaffold will further allow specialization for individual use cases. For instance, an end-to-end RAG pipeline evaluation harness can be built on top of it, implementing a RAG-specific API.

Related to #7415.