Weighted Data Batcher with Checkpointing

Requirements The provided DataRegistry in Colab exposes an interface similar to: Implement the following batcher: Part 1 — Weighted sampling Make next_batch() return exactly batch_size examples, sampled from the registry’s datasets according to weights. For this initial version, you may assume that batch_size is divisible by sum(weights). Batch ordering affects deterministic reproducibility, so specify the ordering convention you use. Part 2 — Deterministic save/resume…