O(1) slot allocation in ReqToTokenPool.alloc() (#32208)

Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>

python/sglang/srt/mem_cache/memory_pool.py

@@ -306,8 +306,14 @@ class ReqToTokenPool:
         need_size = len(reqs) - len(reusing)
         if need_size > len(self.free_slots):
             return None
-        select_index = self.free_slots[:need_size]
-        self.free_slots = self.free_slots[need_size:]
+        if need_size > 0:
+            # Pop from the tail: O(need_size), unlike a prefix pop which is
+            # O(len(free_slots)).
+            select_index = self.free_slots[-need_size:]
+            del self.free_slots[-need_size:]
+        else:
+            # Handled separately: free_slots[-0:] is the entire list, not [].
+            select_index = []
         offset = 0
         for r in reqs:
             if r.req_pool_idx is None: