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
Signed-off-by:

From: @yangzhenzhang
Reviewed-by: @cristoval,@stsuteng
Signed-off-by: @stsuteng

From: @huangxinjing
Reviewed-by: @zhunaipan,@stsuteng
Signed-off-by: @stsuteng

Signed-off-by: alex-yuyue <yue.yu1@huawei.com>

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

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

Merge pull request !7946 from gziyan/fix_auto_optim_shard

Merge pull request !8063 from yihuaijie/dev