* "add nccl enforce"

* Dev

* Update comment

* Add nccl test

* Follow comments

By shell command

```bash
sed -i 's#ifdef PADDLE_ONLY_CPU#ifndef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
sed -i 's#ifndef PADDLE_ONLY_CPU#ifdef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
```

* If stat is an integer, bool value will implicit cast to int before
  pass to PADDLE_ENFORCE

Note: If no symbol with a suitable value is found, both this field and dli_saddr shall be set to NULL.

It is buggy in some Linux because the unique_ptr will be free however
the std::string trying to use that char*.

Moreover, it is no need to demangle for error log by Paddle.
Just use `c++filt` or other shell utilities to do this.

* Wrong Printf format before

* Make `PADDLE_ENFORCE_EQ` supports custom class, like DDim

* fix some enforce

* remove compatible_type to avoid compile error

* remove shared_ptr

* fix tensor error msg

* Make interface of Operator to `const Scope&`

* Use EnforceNotMet to unify all exception types.

Basically from caffe2::logging.h, but only expose `PADDLE_ENFORCE`
interface.