[Dy2Static] Refactor param_guard logic of @to_static

[Aurelius84]

PR types

Function optimization

PR changes

APIs

Describe

1.PR功能

1. 重构了 param_guard的逻辑@to_static
2. 重点覆盖解决 self.weight与各种框架容器、Python list、dict等容器搭配使用时的问题。

2. 解决方案

无论上层VarBase如何被各种容器（Python的list、dict、ParameterList）包含，最终在转为静态图时，都是以OpDesc为单位的操作的，对应于framework.py中的append_op，因此只需要在append_op进行统一的转Variable处理即可。

为什么不对Attrs dict 处理?
答：因为在append_op处的Attrs是不支持Tensor类型的。前端Python API对于参数为Tensor类型的支持方案是：将Tensor类型的参数映射一个新的Input，如slice中starts对应StartTensor。

Attr支持的类型：

enum AttrType {
  INT = 0;
  FLOAT = 1;
  STRING = 2;
  INTS = 3;
  FLOATS = 4;
  STRINGS = 5;
  BOOLEAN = 6;
  BOOLEANS = 7;
  BLOCK = 8;
  LONG = 9;
  BLOCKS = 10;
  LONGS = 11;
  FLOAT64S = 12;
}

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[zhhsplendid]

LGTM

[Aurelius84]

This will not hurt performance of static graph, because it's only call once in static mode.

[Xreki]

LGTM for the modification of framework.py