[Feature] Parallel task-level review for Momus

[popododo0720]

Problem

When plan files exceed 500+ lines, a single Momus instance reviewing the entire file sequentially takes too long.

Proposal: Task-level Parallel Processing

Split the plan into task units and have multiple Momus instances review them concurrently.

• Going more granular than task-level introduces context-understanding overhead
• Task-level is the optimal granularity for parallelization

Implementation Idea

# Current (Sequential)
Momus → [Entire Plan] → Slow

# Improved (Task-level Parallel)
Momus-1 → [Task 1-3]   ─┐
Momus-2 → [Task 4-6]   ─┼─ Concurrent execution
Momus-3 → [Task 7-9]   ─┤
Momus-4 → [Task 10-12] ─┘
              ↓
        Merge results → Final verdict

Reuse Existing ULW Infrastructure

• ConcurrencyManager (per-model concurrent execution)
• delegate_task(background=true) pattern

Expected Benefits

• Faster review time for large plans
• Linear scaling with number of tasks

[justsisyphus]

Thanks for the feature request! Task-level parallel review for Momus is a great optimization idea. The proposal to leverage existing ConcurrencyManager and delegate_task infrastructure makes sense for handling large plan files efficiently. This is marked as enhancement and ready for implementation planning.