Add task category to dataset card

This PR adds the missing `task_categories` field to the dataset metadata so that ReplayBench is properly categorized as a `text-generation` benchmark on the Hugging Face Hub. No other content changes are needed; the existing README already contains the relevant links, benchmark details, citation, and license.

README.md

@@ -1,14 +1,16 @@
 ---
 language:
-  - en
-  - zh
+- en
+- zh
 license: apache-2.0
 pretty_name: ReplayBench
+task_categories:
+- text-generation
 tags:
-  - benchmark
-  - llm-agents
-  - agent-security
-  - tool-result-replay
+- benchmark
+- llm-agents
+- agent-security
+- tool-result-replay
 ---
 
 # ReplayBench