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)
```

* add infer_value for common math ops
* convert constant bool tensor to bool value

* do not infer value when encounter 0 as division

for while condition, do not unrool if condition is a tensor

Signed-off-by: candanzg <zhangshucheng@huawei.com>

fix DiagPart constant folding issue

fix argmin output type check

fix atan2 doc error

fix remove FusedBatchNorm and its grad

Signed-off-by: candanzg <zhangshucheng@huawei.com>

Signed-off-by: leonwanghui <leon.wanghui@huawei.com>