Print dtype in final aggregated result

[linamy85]

Change Summary

As requested in another PR for easier result inspection.

This will impact the following benchmark:

• gemm_multiple_run
• inference_add
• inference_rmsnorm
• inference_silu_mul
• inference_sigmoid

Sample log

Adding the dtype prefix for aggregated result as followed:

[float4_e2m1fn] Total floating-point ops: 9895604649984, Step Time (median): 16.36, Throughput (median): 605.00 TFLOP / second / device, TotalThroughput (median): 4840.01 TFLOP / second, MFU: 52.45%

[junjieqian]

Thank you very much for the quick change! This is what needed as sometimes users want to collect the aggregated results from the log, while the csv files provide detailed information.