[Doc] Cleanup non-english comments (#125)

Signed-off-by: Roger Wang <hey@rogerw.io>

docs/design/vllm_omni_design.md

@@ -449,15 +449,15 @@ AsyncLLM.add_request()
 ↓
 OutputProcessor.add_request()
 ↓
-RequestState.from_new_request() → 创建请求状态
+RequestState.from_new_request() → create request state
 
-AsyncLLM.__init__() / AsyncLLM.generate() / AsyncLLM.encode() →  创建一个Background loop 持续从EngineCore获取输出
+AsyncLLM.__init__() / AsyncLLM.generate() / AsyncLLM.encode() →  create a background loop that continuously pulls output from EngineCore
 ↓
-OutputProcessor.process_outputs() → 更新状态并处理输出
+OutputProcessor.process_outputs() → update state and process outputs
 ↓
-RequestState.make_request_output() → 转换为最终输出,格式为RequestOutput或者PoolingRequestOutput
+RequestState.make_request_output() → convert to the final output, formatted as RequestOutput or PoolingRequestOutput
 ↓
-RequestOutputCollector.put() → 推送到队列（AsyncLLM）
+RequestOutputCollector.put() → push to the queue (AsyncLLM)
 ```
 
 Need to add implementation for an existing method

examples/offline_inference/qwen2_5_omni/extract_prompts.py

@@ -4,7 +4,7 @@ from typing import Optional
 
 
 def extract_prompt(line: str) -> Optional[str]:
-    # 提取第一个 '|' 与第二个 '|' 之间的内容
+    # Extract the content between the first '|' and the second '|'
     i = line.find("|")
     if i == -1:
         return None

vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py

@@ -519,9 +519,9 @@ class Qwen3OmniMoeForConditionalGeneration(
                         tts_eos_thinker=tts_eos_thinker,
                         tts_pad_thinker=tts_pad_thinker,
                     )
-                    # req_embeds: [S,D], req_input_ids: [S]  →  map到当前 span
+                    # req_embeds: [S,D], req_input_ids: [S]  →  map onto the current span
                     seg_len = min(span_len, req_embeds.shape[0])
-                    # 强制对齐 dtype/device/内存布局，避免精度或布局差异导致比较不相等
+                    # Force-align dtype/device/memory layout to avoid mismatches from precision or layout differences
                     inputs_embeds[s : s + seg_len] = (
                         req_embeds[:seg_len].to(device=inputs_embeds.device, dtype=inputs_embeds.dtype).contiguous()
                     )