Benchmarking : Absolute -> Relative imports

[varun-sundar-rabindranath]

SUMMARY:
Fix benchmarking imports - Absolute -> Relative.

TEST PLAN:

• local testing : invoke the benchmark script from outside the repository root.

vllm-test) varun@floppy-fan:~/code$ python3 -m  neuralmagic-vllm.neuralmagic.benchmarks.run_benchmarks --help
usage: run_benchmarks.py [-h] -i INPUT_CONFIG_FILE -o OUTPUT_DIRECTORY

Runs benchmark-scripts as a subprocess

options:
  -h, --help            show this help message and exit
  -i INPUT_CONFIG_FILE, --input-config-file INPUT_CONFIG_FILE
                        Path to the input config file describing the benhmarks to run
  -o OUTPUT_DIRECTORY, --output-directory OUTPUT_DIRECTORY
                        Path to a directory that is the output store

• Run nm-benchmark GHA job.

[andy-neuma]

cool