From: @lianliguang
Reviewed-by: @kisnwang,@zhoufeng54,@chujinjin
Signed-off-by: @chujinjin

Change backend execution order sorting policy from DFS to BFS
in KernelGraph::SetExecOrderByDefault().

The basic idea is: exploits data dependency to control the execution order
of side-effect operations, and keep the semantics of ANF unchanged.

The ControlDepend primitive is removed and there are two primitives added:

1. UpdateState:
```
  a = Assign(para, value)
```
became:
```
  a = Assign(para, value, u)
  u = UpdateState(u, a)
```

2. Load:
```
  x = Add(para, value)
```
became:
```
  p = Load(para, u)
  x = Add(p, value)
  u = UpdateState(u, p)
```

From: @wangnan39
Reviewed-by: @kingxian
Signed-off-by:

From: @louie5
Reviewed-by: 
Signed-off-by:

From: @wangnan39
Reviewed-by: 
Signed-off-by:

From: @liubuyu
Reviewed-by: @kisnwang,@zhoufeng54
Signed-off-by: @zhoufeng54

From: @HulkTang
Reviewed-by: @chujinjin,@kisnwang
Signed-off-by: @chujinjin

From: @irmo
Reviewed-by: @kisnwang,@zhoufeng54
Signed-off-by: @zhoufeng54

From: @jjfeing
Reviewed-by: @zhoufeng54,@chujinjin
Signed-off-by: @chujinjin

From: @dayschan
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @gaoxiong1

From: @yuchaojie
Reviewed-by: 
Signed-off-by:

From: @r1chardf1d0
Reviewed-by: 
Signed-off-by:

From: @dayschan
Reviewed-by: @gaoxiong1,@dylangeng
Signed-off-by: @gaoxiong1

1. added a pass to replace Assign with InplaceAssign.
2. bugfix in eliminate_redundant_output. the side-effect node should not be eliminated.
3. bugfix in graph_kernel/splitter.py, the kernel includes InplaceAssign should be a composite node.
4. added two tool functions GetAllInputDeviceTypes and GetAllOutputDeviceTypes into AnfAlgo.
5. do not fuse a single Assign in pass BasicOpsFusion.

From: @liubuyu
Reviewed-by: 
Signed-off-by:

From: @yuchaojie
Reviewed-by: @kingxian,@zhoufeng54
Signed-off-by: @kingxian

From: @HulkTang
Reviewed-by: 
Signed-off-by:

From: @louie5
Reviewed-by: 
Signed-off-by:

From: @wilfchen
Reviewed-by: @cristoval,@limingqi107
Signed-off-by: @limingqi107

From: @dayschan
Reviewed-by: @gaoxiong1,@ckey_dou
Signed-off-by: @ckey_dou

Cast the float16-input to float32 before ReduceSum, and cast back to float16 after ReduceSum.
If the op after this ReduceSum is a casting from float16 to float32, then it can be eliminated.

From: @jjfeing
Reviewed-by: @zhoufeng54,@kisnwang
Signed-off-by: @kisnwang

From: @liubuyu
Reviewed-by: @zhoufeng54,@jjfeing
Signed-off-by: @jjfeing

From: @yanghaitao1
Reviewed-by: @wangyue01,@lilongfei15
Signed-off-by: @lilongfei15