A queue-based ML pipeline with scikit-learn and Redis workers

Heavy model fitting has no business running on the same event loop as interactive requests.

Bounding the cost

Each job is bounded by input caps and representative sampling, so runtime stays predictable regardless of dataset size.