cherrypick for #32640 :add profile and fix dataset hang in heterps;test=develop

Author: danleifeng

python/paddle/distributed/fleet/dataset/dataset.py

@@ -541,7 +541,7 @@ def _dynamic_adjust_before_train(self, thread_num):
             if self.use_ps_gpu:
                 self.dataset.dynamic_adjust_channel_num(thread_num, True)
             else:
-                self.dataset.dynamic_adjust_channel_num(thread_num, False) 
+                self.dataset.dynamic_adjust_channel_num(thread_num, False)
         self.dataset.dynamic_adjust_readers_num(thread_num)
 
     def _dynamic_adjust_after_train(self):

python/paddle/fluid/dataset.py

@@ -403,7 +403,7 @@ def _dynamic_adjust_before_train(self, thread_num):
             if self.use_ps_gpu:
                 self.dataset.dynamic_adjust_channel_num(thread_num, True)
             else:
-                self.dataset.dynamic_adjust_channel_num(thread_num, False)            
+                self.dataset.dynamic_adjust_channel_num(thread_num, False)
         self.dataset.dynamic_adjust_readers_num(thread_num)
 
     @deprecated(
@@ -415,7 +415,7 @@ def _dynamic_adjust_after_train(self):
             if self.use_ps_gpu:
                 self.dataset.dynamic_adjust_channel_num(self.thread_num, True)
             else:
-                self.dataset.dynamic_adjust_channel_num(self.thread_num, False)        
+                self.dataset.dynamic_adjust_channel_num(self.thread_num, False)
         self.dataset.dynamic_adjust_readers_num(self.thread_num)
 
     @deprecated(