Add task category and code link

This PR adds the `text-retrieval` task category to the dataset metadata and links the dataset card to its GitHub repository. This makes the dataset discoverable when filtering for retrieval tasks and provides the code for reproducing the benchmark.

README.md

@@ -1,4 +1,6 @@
 ---
+task_categories:
+- text-retrieval
 dataset_info:
 - config_name: beamSplitterPlate
   features:
@@ -495,6 +497,8 @@ World Embedding Optics Retrieval is the optics and electromagnetism retrieval sp
 2,700 simulation videos from 27 physics families. The family shards are loaded
 together as the default configuration.
 
+Code: https://github.com/World-Representation-Lab/World-Embedding-Benchmark
+
 ## Usage
 
 ```python
@@ -528,4 +532,4 @@ The benchmark uses `parsed_text` for text-video retrieval.
   primaryClass  = {cs.CV},
   url           = {https://arxiv.org/abs/2610.03632}
 }
-```
+```