Update analyzers to use the new trace call stack graph implementation

[fengxizhou]

🚀 Motivation and context

With the import of the new trace call stack graph implementation, we would like to switch existing client code to use the new implementation to minimize code maintenance burden. With the new implementation, three existing unit tests have failed. We need to modify the client code and the new implementation to pass all tests.

Description

We would to change the following analyzers to use the new trace call stack implementation and deprecate existing call stack implementation. The following changes are required:

• Refactor CuptiCounterAnalysis to use the new Trace CallGraph implementation.
• Update CriticalPathAnalysis to use the new implementation.
• Update get_frequent_cuda_kernel_sequences to use the new implementation.
• Add back CallStackGraph.dfs_traverse() method.
• Correct the assertion on the value of botton_level_op in test_get_cupti_counter_data_with_operators().
• Update infer_device_type() to use pid and tid columns in addition to the stream column.

Alternatives

No response

Additional context

No response

[fengxizhou]

Closed as completed.