[Feature] Align ascend profiler with torch profiler

[DoorKickers]

DIPU ascend profiler对齐

Summary

将DIPU ascend profiler和torch profiler对齐

实现思路

• cpp层通过实现libkineto接口，调用aclprof api，在临时路径生成包含硬件统计信息的ascend PROF文件，并通过libkineto将路径传递到python层
• python层mock了torch profile中的export_chrome_trace方法，加入msprof工具解析、chrome_json合并操作

测试

• 已在910机器上验证过单机单卡和单机多卡

示例用法和结果

output_path = './example.json'
input = torch.randn(2, 3).cuda()
with profile(
    activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
    with_stack=True,
    schedule=torch.profiler.schedule(wait=0, warmup=0, active=2, repeat=2)
) as prof:
    for i in range(10):
      y = fn(input)
      z = y + y    
      prof.step()
prof.export_chrome_trace(output_path)

[CLAassistant]

All committers have signed the CLA.