Replace various key=value string parsers with a single utility

[sjmonson]

Summary

Replaces various config string handlers with a new ArgString format.

Details

Replaces the dataset config parser with a more advance "arg_string" parser with support for nested and list config structures. For example:

prefix_buckets[0].bucket_weight=10,prefix_buckets[0].prefix_count=10,prefix_buckets[1].bucket_weight=20,prefix_buckets[1].prefix_count=4

Renders to

{
    "prefix_buckets": [
        {"bucket_weight": 10, "prefix_count": 10},
        {"bucket_weight": 20, "prefix_count": 4},
     ]
}

Also added some glue logic to allow this format anywhere on the CLI that supports JSON. This code should be considered temporary as it will mostly be replaced through the rest of the CLI refactor.

Also open to suggestions for a better name then "arg_string"; a 2-4 letter acronym like JSON or YAML would be good.

Test Plan

The following benchmark should run without errors and have ~384 ISL per request (make sure to check startup logs confirm backend-kwargs and warmup are configured):

guidellm benchmark run \
    --profile concurrent \
    --data "prompt_tokens=256,output_tokens=256,prefix_buckets[0].bucket_weight=10,prefix_buckets[0].prefix_tokens=128,prefix_buckets[0].prefix_count=10" \
    --request-format /v1/completions \
    --backend-kwargs '{"timeout":120}' \
    --warmup percent=0.1,value=20 \
    --max-seconds 30 \
    --rate 30

---

• "I certify that all code in this PR is my own, except as noted below."

Use of AI

• Includes AI-assisted code completion
• Includes code generated by an AI application
• Includes AI-generated tests (NOTE: AI written tests should have a docstring that includes ## WRITTEN BY AI ##)

[dbutenhof]

Minor comments from a first pass without getting deep into the algorithms ... but it looks good.

[jaredoconnell]

Looks good to me. And basic data args are working for me.

[jaredoconnell]

A test is failing. It looks like maybe you changed the error message under that condition.

[dbutenhof]

This looks great. Minor issue: you added the external get/set methods in the arg string parser class, which aren't used, and aren't tested. If we're going to have them, they should be tested.