[RFC]: send_object_list 和 recv_object_list功能支持

### 需求和背景 在分布式深度学习训练中，进程间通信是核心功能。当前的 MindSpore 分布式通信 API 主要支持 Tensor 数据类型的通信，但缺乏对 Python 对象列表的直接支持。用户需要手动序列化对象、使用底层 Tensor 通信 API，然后手动反序列化，这增加了使用复杂度。 在PP并行动态shape场景，需要把shape的type和layout信息发送给其他卡进行星系传递，该信息为一个python对象，业界模型使用更直观的 API接口：`send_object_list` 和 `recv_object_list` 接口直接发送，简化分布式训练中 Python 对象交换的过程，更直观。 ### 目标 提供简单直观的 API 用于 Python 对象列表的分布式通信 ## 设计方案 ### API 设计 接口名和参数完全对标业界： #### send_object_list ```python def send_object_list(object_list, dst, group=None, device=None): """ Send picklable objects to dst synchronously. Args: object_list (List[Any]): List of input objects to sent. Each object must be picklable. Receiver must provide lists of equal sizes. dst (int): Destination global rank to send object_list to. group (str, optional): The process group to work on. If None, the default process group will be used. Default: None. device (str, optional): Currently it is a reserved parameter. Default: None. Raises: TypeError: If the dst is not an int or group is not a str. ValueError: If the dst process rank id is same as the current process. """ ``` #### recv_object_list ```python def recv_object_list(object_list, src=0, group=None, device=None): """ Receive picklable objects from src synchronously. Args: object_list (List[Any]): List of objects to receive into. Must provide a list of sizes equal to the size of the list being sent. src (int, optional): A required integer identifying the source global rank. Default: 0. group (str, optional): The process group to work on. If None, the default process group will be used. Default: None. device (str, optional): Currently it is a reserved parameter. Default: None. Returns: int, If success, return 0. Raises: TypeError: If the src is not an int or group is not a str. """ ``` ### 实现架构 #### 核心组件 1. **序列化/反序列化模块**： - `_object_to_tensor()`: 将 …