[Doc] Update max_tokens to max_completion_tokens in all docs (#6248)

### What this PR does / why we need it?

Fix:

```
DeprecationWarning: max_tokens is deprecated in favor of the max_completion_tokens field.
```

- vLLM version: v0.14.1
- vLLM main:
d68209402d

Signed-off-by: shen-shanshan <467638484@qq.com>

docs/source/tutorials/Qwen-VL-Dense.md

@@ -142,7 +142,7 @@ llm = LLM(
 )
 
 sampling_params = SamplingParams(
-    max_tokens=512
+    max_completion_tokens=512
 )
 
 image_messages = [
@@ -238,7 +238,7 @@ llm = LLM(
 )
 
 sampling_params = SamplingParams(
-    max_tokens=512
+    max_completion_tokens=512
 )
 
 image_messages = [