Benchmark options

[Kern--]

Issue #, if available:
N/A

Description of changes:

All of our existing benchmark images are run with the same set of containerd
options which are only configurable at the test level to control things
like which snapshotter is used.

This is a problem for benchmarking GPU workloads, for example, where we need to
pass additional options to mount the GPU in the container which don't
apply to all images in the test.

Additionally, our benchmarker assumes that the benchmarked images
require no configuration, however this can make experimentation hard in
cases where a single base-image can be used for multiple use cases
depending on environment variables, confiration mounts, etc.

This change adds the ability to configure image-specific options when
loading benchmarks from json. The options are not required and if not
passed, the benchmarker will behave as it did before this change.

The set of options available in this change are those that were
necessary for benchmarking the LLM workloads that I was trying to test.
They are not comprehensive, but can be built upon as use cases arise.

---

I kept 2 commits to separate some refactoring that I needed to make these change from the changes themselves. I'm happy to squash or pull out the refactoring into a separate commit if that's preferred.

Testing performed:
~/benchmark.json

[{
  "short_name": "djl-serving",
  "image_ref": "<ACCOUNT>.dkr.ecr.us-west-2.amazonaws.com/djl-serving:0.25.0-deepspeed",
  "soci_index_manifest_ref": "<INDEX DIGEST>",
  "ready_line": "ModelServer BOTH API bind to: http://0.0.0.0:8080/",
  "timeout_sec": 1000,
  "options": {
    "net": "host",
    "gpu": true,
    "shm_size": 1000000,
    "mounts": [
      {
        "source": "/home/ubuntu/model",
        "destination": "/opt/ml/model",
        "type": "bind",
        "options": ["rbind", "ro"]
      },
      {
        "source": "/home/ubuntu/model_logs",
        "destination": "/opt/djl/logs",
        "type": "bind",
        "options": ["rbind", "rw"]
      }
    ]
  }
}]

BENCHMARK_FLAGS="-count 1 -f ~benchmark-tgi.json" make benchmarks-perf-test
BENCHMARK_FLAGS="-count 1 -f ~/benchmark-tgi.json" make benchmarks-perf-test

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[sondavidb]

Unrelated, but is there a reason benchmarkTests.go is in camelcase and everything else uses underscores?

[Kern--]

Unrelated, but is there a reason benchmarkTests.go is in camelcase and everything else uses underscores?

No, I don't think so. Historical accident.