Add BALANCED_ASSIGNMENT variable config

[rasmi]

Adds a new BALANCED_ASSIGNMENT variable config type that allows experimenters to randomly assign participants to different experimental conditions (e.g., policies, treatments) with balanced distribution.

Changes

New Feature - Balanced Assignment Variables

• New BalancedAssignmentVariableConfig type with three balance strategies:
  • ROUND_ROBIN - Cycles through values based on participant count (deterministic, perfectly balanced)
  • LEAST_USED - Assigns to the value used by fewest participants (handles dropouts)
  • RANDOM - Seeded random selection without balancing
• Balance scope options: EXPERIMENT-wide or per-COHORT
• Full UI support in the variable editor
• Example usage in policy template (getPolicyBalancedExperimentTemplate)

Refactor - Moved generateVariablesForScope to functions

• Moved from utils/ to functions/ since balanced assignment requires Firestore queries (e.g. querying active participant count for ROUND_ROBIN strategy, variable count for LEAST_USED strategy)
• Now async to support database-dependent variable generation
• Helper functions (generateStaticVariables, generateRandomPermutationVariables) remain in utils and are exported

Testing

• Added integration tests for generateVariablesForScope in functions/ using Firestore emulator
• Updated unit tests in utils/ to test helper functions directly