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

From: @tina_mengting_zhang
Reviewed-by: @john_tzanakakis,@pandoublefeng
Signed-off-by: @john_tzanakakis

dump hccl and config json files to dir

update filename for sync dump except cpu dump

update testcases

From: @kisnwang
Reviewed-by: @limingqi107,@zhoufeng54,@jjfeing
Signed-off-by: @jjfeing

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

1. collect backend graph and graph execution order
2. add trigger point when CPUKernelRuntime::Run() raise "Launch kernel failed"

From: @ding_fei_fei
Reviewed-by: @kingxian,@kingxian
Signed-off-by: @kingxian,@kingxian

From: @zhangbuxue
Reviewed-by: @guoqi1024,@guoqi1024,@zhaizhiqiang
Signed-off-by: @zhaizhiqiang

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

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

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

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

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

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

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

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