Add CPU affinity setting to latency benchmark

[hubertlu-tw]

Motivation

This PR adds CPU affinity setting (i.e. NUMA binding) to latency test (bench_one_batch.py) to increase performance.

Modifications

Leverage "set_gpu_proc_affinity" when SGLANG_SET_CPU_AFFINITY=1

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling.

CC: @HaiShaw

[HaiShaw]

@hubertlu-tw May you provide offline benchmark numbers to two models and two different builds (machines) for the difference? Previously we only enabled this for online cases. Thanks!

[HaiShaw]

@hubertlu-tw Can you please fix the CI/Lint?

[hubertlu-tw]

@hubertlu-tw May you provide offline benchmark numbers to two models and two different builds (machines) for the difference? Previously we only enabled this for online cases. Thanks!

I tested the changes on two different models (DeepSeek-V3 and Llama-3.1-70B) and two different servers (with different system setups). I have consistently observed 1-5% perf improvement with CPU affinity setting while running bench_one_batch.

@HaiShaw the Lint error is resolved. Thanks.