(opt) lock free DAG

[AshinGau]

Performance Optimizations: Lock-free DAG & Scheduling Improvements

This PR introduces significant performance optimizations to the DAG execution engine:

Lock-free DAG Dependency Updates

• Eliminated mutex contention during dependency updates using lock-free DAG implementation
• Enables fully concurrent dependency resolution

​Atomic-based Scheduler

• Replaced traditional locks with AtomicUsize-based cursor for scheduling
• Reduces synchronization overhead in task dispatching

​Priority Execution Optimization

• Accelerates scheduling of distance-1 execution tasks through fast-path detection
• Minimizes latency for immediate dependency resolution

Lock-free Continuity Detection

• Implemented non-blocking ContinuousDetectSet using atomic operations
• Ensures zero-wait detection of execution continuity

Performance Impact:

These optimizations collectively achieve ​over 30% throughput improvement in task execution workloads, particularly benefiting highly parallel dependency graphs.