【Pull Request】Adapting to a new Torch profiling data

Thank you very much for developing this tool, which is a great convenience for the visualization of model performance tuning. We have optimization data collected by the torch framework running based on the new AI chip architecture. This data format is different from xxx.pt.trace.json. It is a directory. However, the views displayed on the Tensorboard is the same. Therefore, some data adaptations are performed, including:
1. Verify the file/directory name rule.
2. Some new data reading methods are added. (In addition to JSON files, CSV files are read.)
3. Modified some functions and classes.
The preceding operations are only compatible with the new data format and have no impact on the original data parsing mode. Currently, the fronted web pages use Google Charts, which depends on the network. To ensure that tool can be used offline, the following modifications are made:
1. Replaced Google Charts with echarts. (Echarts line charts have better performance in big data scenarios.)
2. In the new data format, some views (such as Kernel and Memory) are modified, but the original data display is not affected.
3. The webcomponents.min.js file imported in trace_embedding.html is downloaded offline as a local static file.
Do you still accept pull request? The code changes are about 2k lines.

Start with the existing handling of xxx.pt.trace.json and the frontend entry point trace_embedding.html. Compare the new directory and CSV profiling format with the original parsing and review the proposed Google Charts, echarts, and local webcomponents.min.js changes. Done means the new data and offline views work without affecting original data display.