FEAT Support baseline-only execution in Scenario

[slister1001]

Summary

This PR enables scenarios to be initialized with an empty strategies list when include_baseline=True. Previously, empty strategies would raise a ValueError even when baseline was requested.

Problem

When using RedTeamAgent (or FoundryScenario) with include_baseline=True and scenario_strategies=[], the scenario would fail with:

ValueError: No valid FoundryStrategy strategies provided.

This prevented users from running baseline-only attacks, which is useful for:

• Establishing baseline metrics before applying attack techniques
• Quick sanity checks of target behavior
• Comparing raw objective success rates against attack-modified prompts

Changes

1. scenario_strategy.py: Added allow_empty parameter to prepare_scenario_strategies(). When True and an empty sequence is explicitly provided, returns an empty list instead of raising ValueError.

2. scenario.py:

   • Updated initialize_async() to pass allow_empty=self._include_baseline so baseline-only execution is allowed when baseline is requested
   • Added _create_standalone_baseline() method that creates a baseline attack directly from dataset_config when no other atomic attacks exist to derive from

3. Tests: Added 4 unit tests covering baseline-only execution scenarios

Usage Example

scenario = RedTeamAgent(
    include_baseline=True,
    attack_scoring_config=scoring_config,
)

await scenario.initialize_async(
    objective_target=target,
    scenario_strategies=[],  # Empty - baseline only
    dataset_config=dataset_config,
)

result = await scenario.run_async()
# Result contains only baseline attack results

Test Results

All 29 scenario tests pass, including 4 new tests for baseline-only execution.

[rlundeen2]

This is good!

Can we get rid of the if statement and just get scenario-level settings? The only reason we didn't do that before was these didn't exist.