fix: use cache_salt for gpt-oss (#23186)

Signed-off-by: Marko Rosenmueller <5467316+dr75@users.noreply.github.com>

vllm/entrypoints/openai/serving_chat.py

@@ -1483,4 +1483,9 @@ class OpenAIServingChat(OpenAIServing):
         # Render prompt token ids.
         prompt_token_ids = render_for_completion(messages)
         engine_prompt = EngineTokensPrompt(prompt_token_ids=prompt_token_ids)
+
+        # Add cache_salt if provided in the request
+        if request.cache_salt is not None:
+            engine_prompt["cache_salt"] = request.cache_salt
+
         return messages, [prompt_token_ids], [engine_prompt]