Move instances under tasks directory

Change the dataset layout to tasks/<instance-id>/ without modifying instance contents; update the dataset card structure documentation.

README.md

@@ -60,8 +60,8 @@ source provenance, and generation seeds.
 Two instances were generated with **seed 2**, while their directory names
 retain the dataset-wide `__seed31415` suffix:
 
-- `computer_science.deployment_prediction_sets__seed31415`
-- `math.ucb_q_learning_regret__seed31415`
+- `tasks/computer_science.deployment_prediction_sets__seed31415`
+- `tasks/math.ucb_q_learning_regret__seed31415`
 
 Their actual seed is recorded in `metadata.json` under `seed_overrides`. Folder
 suffixes must not be used as the sole source of generation provenance.
@@ -90,19 +90,24 @@ included in this published dataset.
 ## Repository structure
 
 This is not a conventional row-oriented dataset and does not define standard
-train, validation, or test splits. Each top-level task directory is one
-generated benchmark instance:
+train, validation, or test splits. Root-level metadata is separated from the
+generated benchmark instances under `tasks/`:
 
 ```text
-<domain>.<task_id>__seed31415/
-├── prompt_b1.md
-├── prompt_b2.md
-├── prompt_b3.md
-├── prompt_b4.md
-├── data/
-├── reference/
-├── instance_meta.json
-└── framework_task_info.json
+.
+├── README.md
+├── metadata.json
+├── .gitattributes
+└── tasks/
+    └── <domain>.<task_id>__seed31415/
+        ├── prompt_b1.md
+        ├── prompt_b2.md
+        ├── prompt_b3.md
+        ├── prompt_b4.md
+        ├── data/
+        ├── reference/
+        ├── instance_meta.json
+        └── framework_task_info.json
 ```
 
 - `prompt_b1.md` through `prompt_b4.md` are four disclosure levels of the task