AutoUnit non_blocking=True (#886)

Summary:
Pull Request resolved: #886

# Context
When adding flag which disabled prefetch from AutoUnit, I made the default `non_blocking=False` when moving batch from cpu->gpu.

User here: https://fb.workplace.com/groups/1323951304836028/permalink/1718731732024648/ makes good point to make these async

# This DIff
Changes to `non_blocking=True`

Reviewed By: diego-urgell

Differential Revision: D61831174

fbshipit-source-id: 6f522f78b9af9cc6217b2117916ec8cd76e23365

Author: JKSenthil

torchtnt/framework/auto_unit.py

@@ -259,7 +259,7 @@ def _prefetch_next_batch(self, state: State, data_iter: Iterator[TData]) -> None
     def _get_next_batch(self, state: State, data: Iterator[TData]) -> TData:
         if not self._enable_prefetch:
             batch = next(data)
-            return self.move_data_to_device(state, batch, non_blocking=False)
+            return self.move_data_to_device(state, batch, non_blocking=True)
 
         active_phase = state.active_phase
         if not self._phase_to_prefetched[active_phase]: