bench config: ship Ingot-Bench as Parquet so load_dataset(..., "bench") works (JSONL kept for the grader)

README.md

@@ -27,7 +27,7 @@ configs:
 - config_name: repo
   data_files: data/repo/*.parquet
 - config_name: bench
-  data_files: bench/ingot_bench_v1.jsonl
+  data_files: bench/ingot_bench_v1.parquet
 ---
 
 # Ingot (v1.2)
@@ -53,6 +53,8 @@ from datasets import load_dataset
 ds = load_dataset("ego0op/ingot", "tested", split="train")
 ```
 
+The bench ships as both `bench/ingot_bench_v1.parquet` (the `bench` config) and `bench/ingot_bench_v1.jsonl`. The JSONL is what `tools/bench_grade.py` reads.
+
 ## Row schema
 
 | field | meaning |

bench/ingot_bench_v1.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1173f93991fb33ff10160c431a49bb0d7c7dc23bb360aca019fdfcfc16cb6014
+size 3080947