Build a versioned DVC pipeline (dvc.yaml) that chains data prep, featurization, and training stages with cached, reproducible runs

Steps

1. Track raw input data with `dvc add`
2. Create each stage with `dvc stage add -n <name> -p <params> -d <deps> -o <outs> <command>`, chaining outputs of one stage as dependencies of the next
3. Commit the generated `dvc.yaml` (and `.gitignore` entries) to Git
4. Run `dvc repro` to execute the pipeline; DVC skips stages whose dependencies/params haven't changed
5. Commit the resulting `dvc.lock` file, which records hashes of deps, params, and outputs