feat(reporting): add optional kernel summary to perf report

[poznano-amd]

When looking at perf reports of vLLM workloads, we noticed it was challenging to understand which GPU kernels were the most important. This was due to the fact that our cpu_ops had a list of GPU kernels, and we were not able to easily view them in the spreadsheet.

Therefore, we added an optional flag:

  --enable_kernel_summary
                        Enable kernel summary sheet in the report. Disabled by default.

This new kernel_summary sheet is has a list of kernels sorted by the % of total time.

Here's an example spreadsheet: [available by request]