only Linux system is supported now.

change the default value of `ENABLE_AKG` to off, and controlled by option `-K`.
the `ENABLE_AKG` is auto enabled when `ENABLE_GPU` or `ENABLE_D` is on.
since now, we can use `ENABLE_AKG` to control the compilation of graphkernel
and akg codes.

fix usage description for option "-K", it should be "[-K on|off]".

LLVM is required by akg for cpu kernels, so AKG for cpu is default disabled now.

It's necessary to immediately insert cast for the nodes that dtype was changed in SelectKernel,
instead of after BuildKernel. otherwise, some fusion passes cannot work as expected,
like Graph-Kernel optimization.

By now, the shape of LSTM's reserved output will be changed in InitKernel (called by BuildKernel),
and this output is only used by its gradient operator, so we can ignore the reserved output in this pass.

1. Create Tensor and DeviceAddress for output before Launch.
2. Push Launch/Build Task to Queue and execute togather.

This reverts commit b077aa1cab.

Revert "[feat] [assistant] [I3T96X] add new Dataset operator LibriSpeechDataset"

This reverts commit 4e6f7dc97d.

delete pass_registry_test.cc

comment  hiai_nlu_model_multi.pb related  line

Signed-off-by: zjun <zhangjun0@huawei.com>

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

Signed-off-by: zhoufeng <zhoufeng54@huawei.com>

From: @zhangbuxue
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)
```