[RFC]: Mem Tracker重构

### 背景与目标描述. 当前Mem tracker因为历史原因，信息收集点分散于框架运行时的各处，且需要保持顺序 1、上报TaskInfo 2、上报MemInfo 3、上报MemBlockInfo 4、上报UseMemBlock 其中1、2两步分散在运行时各处，在新增内存申请的场景下容易遗漏，导致统计不全的情况  ### 建议的方案. 现在希望将运行时的内存信息收集归一，将原本的TaskInfo和MemInfo结构体信息统一为一个MemInfo，且在内存申请时限制传入完整的MemInfo，在内存申请时增加对MemInfo完整性的校验 `AscendResManager::AllocateMemory(DeviceAddress, stream_id, MemInfo)` MemInfo结构体设计为 ``` struct MemInfo { std::string node_name; std::string graph_name; std::string task_name; int64_t time_stamp; std::string task_file_name; size_t task_line_num; MemType type; size_t size; const void *kernel_tensor(DeviceAddress *); std::string mem_info_file_name; size_t mem_info_line_num; } ``` 当前问题为，代码中原本的调用宏收集信息的行为 ``` device::tracker::CALL_MEMORY_TRACKER_WITH_FILE(AddTask, "PyNative", node->fullname_with_scope(), op_run_info->base_op_run_info.op_name); ``` 会变成往结构体中写入信息，代码观感很差，且该对象需要在运行时内存分配流程中一路向下传递 ``` auto task_info = std::make_shared<MemInfo>(); task_info->task_name = "PyNative"; task_info->node_name = node->fullname_with_scope(); task_info->graph_name = op_run_info->base_op_run_info.op_name; task_info->time_stamp = MemTrackerManager::GetInstance().get_time_stamp(); task_info->task_file_name = FILE_NAME; task_info->task_line_num = __LINE__; ``` ### 期望的反馈时间. ### CC List. ### 其他补充信息. ### Before submitting a new issue... - [x] Make sure you already searched for previous [RFCs](https://gitee.com/mindspore/mindspore/issues?q=is%3Aall+label%3ARFC+sort%3Arecently-updated).