[Bug][DebateWithJudge] run() and batched_run() reject img/imgs, while the telemetry decorator declares it captures them

DebateWithJudge.run() and DebateWithJudge.batched_run() accept no images. Every other multi-agent structure does, and this one's own telemetry decorator declares that it captures img and imgs — so the class advertises image support in three places and implements it in none.

Root cause

swarms/structs/debate_with_judge.py:261 (the decorator) and :264 / :456 (the signatures):

    @trace_run(
        "DebateWithJudge.run",
        input_params=("task", "tasks", "img", "imgs"),
    )
    def run(self, task: str) -> Union[str, List, dict]:

    def batched_run(self, tasks: List[str]) -> List[str]:
        return batched_run(self.run, tasks)

trace_run binds by name and skips anything absent (swarms/telemetry/otel.py:846-850, if p in bound.arguments), so the three phantom parameters are silently dropped rather than raising — which is why nothing has surfaced this.

The images never reach the agents either. Agent.run takes img/imgs, and the three self.*_agent.run(...) calls inside the loop pass neither.

execution_utils.batched_run — the helper DebateWithJudge.batched_run delegates to — already knows how to fan images out per task (img= broadcast, imgs= paired by position, swarms/structs/execution_utils.py:18-20). It is only ever called here with two arguments, so that plumbing is unreachable from this structure.

Signatures across the neighbours:

$ python -c "import inspect; ..."
DebateWithJudge.run       : (self, task: str) -> Union[str, List, dict]
DebateWithJudge.batched_run: (self, tasks: List[str]) -> List[str]
SequentialWorkflow.run    : (self, task: str, img: Optional[str] = None, imgs: Optional[List[str]] = None, *args, **kwargs)
ConcurrentWorkflow.run    : (self, task: str, img: Optional[str] = None, imgs: Optional[List[str]] = None, streaming_callback: ...)

Reproducer

from swarms.structs.debate_with_judge import DebateWithJudge

d = DebateWithJudge(preset_agents=True, max_loops=1, verbose=False)

print("has img param on run? ",
      "img" in DebateWithJudge.run.__wrapped__.__code__.co_varnames)
try:
    d.run("Motion: X", img="chart.png")
except TypeError as e:
    print("run(task, img=...)        ->", type(e).__name__ + ":", e)
try:
    d.batched_run(["a"], imgs=["chart.png"])
except TypeError as e:
    print("batched_run(tasks, imgs=) ->", type(e).__name__ + ":", e)

Output on master @ 444cfc81 (v15.0.1):

has img param on run?  False
run(task, img=...)        -> TypeError: DebateWithJudge.run() got an unexpected keyword argument 'img'
batched_run(tasks, imgs=) -> TypeError: DebateWithJudge.batched_run() got an unexpected keyword argument 'imgs'

So "debate this chart" is not expressible: the only route is to describe the image in the task string and lose it, which is the same defect #1903 describes for HeavySwarm.

Suggested change

    def run(
        self,
        task: str,
        img: Optional[str] = None,
        imgs: Optional[List[str]] = None,
    ) -> Union[str, List, dict]:
        ...
        pro_argument = self.pro_agent.run(
            task=pro_prompt, img=img, imgs=imgs, messages=...
        )

and forward them from batched_run:

    def batched_run(
        self,
        tasks: List[str],
        img: Optional[str] = None,
        imgs: Optional[Sequence[Optional[str]]] = None,
    ) -> List[str]:
        return batched_run(self.run, tasks, img=img, imgs=imgs)

Two decisions worth making explicitly rather than by default:

• Whether the judge sees the image. Giving it to the debaters only is defensible — the judge evaluates arguments, not evidence — but then JUDGE_ROUND_PROMPT should say the image is not shown to it. Giving it to all three is the simpler contract. I lean to all three, matching SequentialWorkflow, where every agent in the chain receives the same img.
• tasks in the decorator. input_params=("task", "tasks", "img", "imgs") names a tasks parameter run will never have — batched_run is a separate method and is not decorated. It should be ("task", "img", "imgs") once the two real ones exist.

Acceptance criteria

• DebateWithJudge().run(task, img=...) and .run(task, imgs=[...]) are accepted and reach the agents.
• batched_run(tasks, imgs=[...]) pairs one image per task, as execution_utils.batched_run already implements.
• trace_run's input_params names only parameters run actually has.

Environment

swarms master @ 444cfc81 (v15.0.1), Python 3.12, macOS.