style(types): fix return type annotation for get_all_requests (#3919)

* style(types): fix return type annotation for get_all_requests

* fix: safely handle empty request lists in RequestManager.step

---------

Co-authored-by: xiongjie <[email protected]>

Author: xiaoajie738

lmdeploy/pytorch/engine/request.py

@@ -225,7 +225,7 @@ def has_requests(self):
             return False
         return not self.requests.empty()
 
-    async def get_all_requests(self) -> Dict[RequestType, Request]:
+    async def get_all_requests(self) -> Dict[RequestType, List[Request]]:
         """Get all requests in current queue."""
         num_reqs = self.requests.qsize()
         reqs: ReqList = []
@@ -247,7 +247,7 @@ def __proc_reqs(elem):
             __proc_reqs(elem)
 
         # gather requests
-        reqs_by_type: Dict[RequestType, Request] = dict((t, []) for t in RequestType)
+        reqs_by_type: Dict[RequestType, List[Request]] = dict((t, []) for t in RequestType)
         for req in reqs:
             reqs_by_type[req.type].append(req)
         return reqs_by_type
@@ -300,11 +300,10 @@ def _log_reqs(reqs: ReqList):
 
         # handle requests
         for req_type in self.request_priority:
-            # request exists
-            if req_type not in reqs_by_type or len(reqs_by_type) == 0:
+            reqs: ReqList = reqs_by_type.get(req_type, [])
+            if not reqs:
                 continue
 
-            reqs: ReqList = reqs_by_type[req_type]
             _log_reqs(reqs)
             self.process_request(req_type, reqs, **kwargs)