BUG: Scorer evaluation accepts zero trials then fails after dataset extraction #3901

Describe the bug

ObjectiveScorerEvaluator.evaluate_dataset_async accepts num_scorer_trials=0, extracts dataset messages through the memory interface, and then fails with an internal NumPy IndexError. Reject nonpositive trial counts at entry, before dataset extraction or scorer calls.

Source: pyrit/score/scorer_evaluation/scorer_evaluator.py:356-436,793-839 at the tested revision.

Steps/Code to Reproduce

import asyncio
from unittest.mock import AsyncMock, MagicMock
from pyrit.models import Message, MessagePiece, Score
from pyrit.score import (
    HumanLabeledDataset, MetricsType, ObjectiveHumanLabeledEntry,
    ObjectiveScorerEvaluator, TrueFalseScorer,
)

async def main():
    scorer = MagicMock(spec=TrueFalseScorer)
    scorer._memory = MagicMock()
    scorer._memory.add_message_to_memory_async = AsyncMock()
    scorer.get_chat_target.return_value = None
    scorer.score_async = AsyncMock(
        return_value=[Score(score_type="true_false", score_value="true")]
    )
    response = Message(message_pieces=[
        MessagePiece(role="assistant", original_value="response")
    ])
    dataset = HumanLabeledDataset(
        name="trial-boundary", version="1", metrics_type=MetricsType.OBJECTIVE,
        entries=[ObjectiveHumanLabeledEntry([response], [True], "objective")],
    )
    await ObjectiveScorerEvaluator(scorer).evaluate_dataset_async(
        dataset, num_scorer_trials=0
    )

asyncio.run(main())

Expected Results

A clear ValueError before memory/scorer calls. Cover zero and negative counts and preserve one-trial metrics. Apply the same contract consistently to higher-level evaluation entry points.

Actual Results

IndexError: too many indices for array: array is 1-dimensional, but 2 were indexed

The original probe records one awaited memory-add call and zero scorer calls at zero trials. A one-trial control succeeds with accuracy 1.0 and trial_scores == [[True]].

Screenshots

N/A.

Versions

Linux; Python 3.12.3; PyRIT 1.2.0.dev0 at 01ce6ca66a315770b36c7dca1f320668c0d92d7c. Actual evaluator/models/metrics, synthetic one-entry dataset, mocked memory and scorer. No real database, CLI, or external inference. Full package-version output was not collected.