From: @jiang-shuqiang
Reviewed-by: 
Signed-off-by:

From: @mind-lh
Reviewed-by: @liangchenghui,@wuxuejian
Signed-off-by: @liangchenghui

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: @TFbunny
Reviewed-by: 
Signed-off-by:

Fix can not collect input data when batch size is 1 and total step
number is 1

Fixed spelling errors

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

From: @lijiaqi0612
Reviewed-by: @youui,@liangchenghui
Signed-off-by: @liangchenghui

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

Merge pull request !2399 from ghzl/fix-params-keyerror-in-group-params

Merge pull request !2140 from ShidaHe/debugger_dev

The summary operator will be optimized when it return the origin value
in constant folding scene. So I return a None value to avoid this.