Support Multiple Model initialization

[gwarmstrong]

This PR supports having multiple models initialized in a generate command, like so:

from nemo_skills.pipeline.cli import wrap_arguments, generate

generate(
    ctx=wrap_arguments(""),
    cluster="ord",
    partition="interactive",
    model=["/hf_models/Qwen3-32B", "/hf_models/Qwen3-8B"],
    server_gpus=8,  # Broadcast to both models
    server_nodes=[2,1], # 2 nodes for 32B, 1 node for 8B
    server_type="vllm",
    generation_module="multiturn-conversation/multiturn_conversation.py",
    input_file="/nemo_run/code/multiturn-conversation/sample_prompts.jsonl",
    output_dir=f"/workspace/multiturn-conversation/minimal_example/output",
)

This generation module is not available in the branch (and likely won't due to the specificity of its implementation) but reach out to me if you want the example.

[Kipok]

lgtm, thanks! Feel free to merge as long as both gpu and slurm tests are passing

[gwarmstrong]

Closed in favor of #1052