Smoke job repo: grouped tabular dataset + config

400 rows / 200 patients, two samples each, so grouped splitting is actually
exercised. Signal is modest and noisy on purpose — a separable toy would score
1.0 with a broken model and prove nothing.

.gitignore

@@ -0,0 +1,19 @@
+# Job outputs — written into the checkout at run time, never committed.
+results/
+logs/
+
+# Large local caches. The provider runs `git clean -fd` WITHOUT -x before each
+# job, so anything ignored here survives between runs on the same box. Put real
+# datasets here; the smoke CSV below is deliberately small enough to commit.
+datasets/
+features/
+models/
+*.h5
+*.ckpt
+*.pt
+*.pth
+
+__pycache__/
+*.py[cod]
+.venv/
+*.env