Refactor CI workflow; Refactor GPU_ubench; Add -f flag to hw stats

[JRPan]

No description provided.

[JRPan]

has dependent on accel-sim/gpu-app-collection#75

[JRPan]

also added -f flag to allow overwriting the NCU profiling output file. There are two steps: generate the output and convert the output to CSV. The first step fails, as the output file already exists. But the second step still runs, generating CSV from the stale old profiling output. So right now, when you run run_hw multiple times, the output is always the same.

Even though I don't know why you would run the script multiple times.