[PHI] Support code auto-gene for Sparse backward C++ API

[zyfncg]

PR types

New features

PR changes

APIs

Describe

Sparse backward API支持通过Yaml配置完成接口代码的自动生成。Sparse 前向API可参考#PR40060

以sparse_conv3d_grad为例，其yaml配置格式如下：

- sparse_bw_api : conv3d_grad
  forward : conv3d (Tensor x, Tensor kernel, int[] paddings, int[] dilations, int[] strides, int groups) -> Tensor(out@SparseCooTensor), Tensor(rulebook@DenseTensor)
  args : (Tensor x, Tensor kernel, Tensor rulebook, Tensor out_grad, int[] paddings, int[] dilations, int[] strides, int groups)
  output : Tensor(x_grad@DenseTensor), Tensor(kernel_grad@DenseTensor)
  kernel :
    func : sparse_conv_grad

配置规则与普通反向API基本相同，区别在于：

1. 无需配置infer_meta项
2. output项中的变量名需使用 "@xxx" 的后缀标记返回Tensor的具体数据类型

[paddle-bot-old]

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