feat: Support a list of concurrencies for bench configs

[micah-wil]

This PR allows us to specify a list of concurrencies for bench configs in workloads. For example:

vllm_bench:
  configs:
    - name: 8k-in-1k-out
      backend: openai
      dataset: random
      input_len: 8192
      output_len: 1024
      num_prompts: 512
      max_concurrency: [1, 64, 128]

The above would run a an 8k/1k benchmark for each concurrency listed (1, 64, 128). You can also optionally change num_prompts to a list to correspond to each max_concurrency, e.g.:

vllm_bench:
  configs:
    - name: 8k-in-1k-out
      backend: openai
      dataset: random
      input_len: 8192
      output_len: 1024
      num_prompts: [10, 64, 1280]
      max_concurrency: [1, 64, 128]

In expand_bench_config, -conc-<value> is appended to each config's name for each conc, so in the above example the final names for the config would be 8k-in-1k-out-conc-1, 8k-in-1k-out-conc-64, 8k-in-1k-out-conc-128.

I've updated workloads/qwen3_5_mi355x.yaml as an initial example.