From: @dinglongwei
Reviewed-by: @wuxuejian,@liangchenghui
Signed-off-by: @wuxuejian

From: @huangbo77
Reviewed-by: @wuxuejian,@oacjiewen
Signed-off-by: @wuxuejian

From: @shibeiji
Reviewed-by: @wuxuejian,@liangchenghui
Signed-off-by: @wuxuejian

From: @he-botao
Reviewed-by: @wuxuejian,@liangchenghui
Signed-off-by:

From: @yuanwei66
Reviewed-by: 
Signed-off-by:

Add L2NormalizeGradCPUKernel 修改测试用例误差判断方式

Add L2NormalizeGradCPUKernel 注释调试信息

Add L2NormalizeGradCPUKernel 修改反向算子的计算公式

Add L2NormalizeGradCPUKernel 去掉 float16 的注册

Add L2NormalizeGradCPUKernel 使用相对误差

Add L2NormalizeGradCPUKernel 更新反向公式

Add L2NormalizeGradCPUKernel 删除调试信息

清除告警

Add L2NormalizeGradCPUKernel 添加测试用例

Add L2NormalizeGradCPUKernel 删除多余的函数

Add L2NormalizeGradCPUKernel 修改注释中的时间

Add L2NormalizeGradCPUKernel 格式化代码

Add L2NormalizeGradCPUKernel 格式化代码，修改 cpplint 问题

Add L2NormalizeGradCPUKernel 修改 cpplint，pylint 问题

Add L2NormalizeGradCPUKernel 修改求导函数，与 GPU 和 Ascend 保持一致。
修改后的公式在数学意义上有问题，但已经和武雪剑对齐，认为没有影响，没有必要要求 GPU 和 Ascend 修改代码。

Add L2NormalizeGradCPUKernel 精简测试用例

From: @xuguoyang5566
Reviewed-by: 
Signed-off-by:

From: @ZhengQihao3f3f3f
Reviewed-by: 
Signed-off-by:

From: @wangyanling10
Reviewed-by: 
Signed-off-by:

From: @zhao_ting_v
Reviewed-by: @wuxuejian
Signed-off-by: @wuxuejian

From: @wuxuejian
Reviewed-by: @kisnwang,@liangchenghui
Signed-off-by: @liangchenghui

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

From: @liu_xiao_93
Reviewed-by: @liangchenghui
Signed-off-by: @liangchenghui

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

From: @wangyanling10
Reviewed-by: 
Signed-off-by:

---
Remove the routine of handling isolated nodes in Renormalize.
Add isolated nodes from Parser&Resolver.
Modify isolated nodes handling in FG&Manager.
Optimize the renormalize routines.
Other optimizations.

From: @yang_chun
Reviewed-by: @wuxuejian,@c_34
Signed-off-by: @wuxuejian

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

From: @yanglf1121
Reviewed-by: 
Signed-off-by:

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: @liangzhibo
Reviewed-by: @ginfung,@zh_qh
Signed-off-by: @zh_qh

From: @wangrao124
Reviewed-by: @wuxuejian
Signed-off-by: