fix: remove outdated Future Improvements section regarding embedding model

## Problem

In the **Kubeflow Pipelines** section of the `README.md`, there is a "Future Improvements" paragraph suggesting that the embedding model should be extracted into a separate service to avoid installing the heavy `sentence-transformers` package during every run.

However, in the final **Modern Infrastructure & CI/CD** section (under Pipeline Optimizations), the README states that the ingestion pipeline was rewritten and the `all-mpnet-base-v2` model is now baked directly into a custom Docker image (`Dockerfile.pipeline`).

This implemented optimization contradicts the older "Future Improvements" suggestion, making it outdated and potentially confusing for readers trying to understand the architectural direction.

## Proposed fix

Remove the outdated "Future Improvements" subsection from the "Kubeflow Pipelines" section as the architectural problem has already been solved using a custom Docker image.

## Acceptance criteria

- The outdated "Future Improvements" section under "Kubeflow Pipelines" is completely removed from `README.md`.
- No conflicting architectural goals remain in the documentation regarding the embedding model optimizations.