Created by: liym27

Please review a small PR splited from This PR : https://github.com/PaddlePaddle/Paddle/pull/22678

1. 本PR支持动态图转静态图：

• 识别：Paddle 动态图API
• 改写：动态图API对应静态图API

2. API改写说明

示例1

• 动态图代码：

@dygraph_to_static_output
def fn_1(input):
        # 该节点将被删除
        pool2d_max = fluid.dygraph.Pool2D(pool_size=2, pool_type='max', pool_stride=1, global_pooling=False)
        
        pool2d_max(input)             #  1.普通表达式。 将被转化
        res_1 = pool2d_max(input) # 2.赋值。将被转化
        res_2 = pool2d_max(input) + pool2d_max(input) # 3. ast子节点调用动态图接口。将被转化

• 转化后的静态图：

def fn_1(input):
        #  1
        fluid.layers.pool2d(input, pool_size=2, pool_type='max', pool_stride=1, global_pooling=False)
        
        # 2
        res_1 =  fluid.layers.pool2d(input, pool_size=2, pool_type='max', pool_stride=1, global_pooling=False)
 
         # 3
        res_2 = fluid.layers.pool2d(input, pool_size=2, pool_type='max', pool_stride=1, global_pooling=False) + fluid.layers.pool2d(input, pool_size=2, pool_type='max', pool_stride=1, global_pooling=False)    

示例2

• 动态图

class DygraphLayer(fluid.dygraph.Layer):
    def __init__(self):
        super(DygraphLayer, self).__init__()
        self.fc = fluid.dygraph.nn.Linear(
            input_dim=10,
            output_dim=5,
            act='relu',
            param_attr=fluid.ParamAttr(initializer=fluid.initializer.Constant(
                value=0.99)),
            bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Constant(
                value=0.5)), )

    def forward(self, inputs):
        prediction = self.fc(inputs)
        return prediction

• 转化后的静态图：

def DygraphLayer_forward(inputs):
      prediction = fluid.layers.fc(input, size=5, act='relu', param_attr=fluid.
        ParamAttr(initializer=fluid.initializer.Constant(value=0.99)),
        bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Constant(
        value=0.5)), num_flatten_dims=-1)
        return prediction

todo

• 完善所有基本api的测试和转化 https://github.com/PaddlePaddle/Paddle/pull/22678
• 自定义类的转化[进行中]

• 本PR分成小PR