Update conv_transpose api

[LielinJiang]

PR types

New features

PR changes

APIs

Describe

Update conv_transpose2d, conv_transpose3d, ConvTranspose2d, ConvTranspose3d api

TODO:

1. 支持output_size为tensor的输入（静态图）。
2. 经过粗略的测试与竞品参考（tf），目前use_cudnn在使用depthwise默认会设置为false，并且使用depthwise_conv_transpose op。后续可能任然需要更多的测试。部分测试数据如下：
   
   
   

[paddle-bot-old]

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

[willthefrog]

for future reference, maybe add the reasoning and bench result of use_cudnn setting to PR description

also, the code does not seem to be DRY enough, consider:

• extract conv_nd function
• reference to class description in functional docstring

[willthefrog]

need subsequent PR to add tensor input for output_size, otherwise LGTM