feat(engine): add parallel execution support for iteration nodes

[cumthxy]

Summary

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Refactoring

Related Issue

Changes

• Add parallel execution support for iteration nodes with configurable concurrency, timeout, and error handling strategies
• Isolate child iteration execution context, including callback queues, variable pools, and history state, to prevent cross-batch interference
• Improve workflow engine task handling and cleanup to better support iteration subgraph execution and cancellation scenarios
• Add iteration sub-DSL extraction and chain resolution enhancements for standalone child workflow execution
• Update ordered streaming callback handling to keep parallel iteration outputs isolated and correctly sequenced

Testing

• Existing tests pass
• New tests added (if applicable)
• Manual testing completed

Screenshots (if applicable)

Checklist

• Code follows project coding standards
• Self-review completed
• Documentation updated (if needed)
• Breaking changes documented

[gemini-code-assist]

Code Review

This pull request introduces parallel execution capabilities for iteration nodes in the workflow engine. Key changes include the implementation of a parallel execution mode with configurable concurrency and error strategies (fail_fast, continue, ignore_error_output), the addition of isolated child engines for parallel batches, and deep copying of variable pools to prevent state sharing. The engine now also supports ordered event relaying from child iterations. Feedback highlights potential improvements in exception handling within the event relay logic and suggests optimizing task management in the queue consumption loop to avoid unnecessary task recreation.

[gemini-code-assist]

Catching a generic Exception here is too broad and might mask unexpected bugs. It is better to catch specific exceptions that are expected from done_task.result(), such as asyncio.CancelledError or specific errors raised by the child engine.

[gemini-code-assist]

Creating new tasks in every iteration of the loop for child_stream_queue.get() and child_order_queue.get() is inefficient. If one task completes, the other is cancelled and recreated in the next iteration. Consider reusing the pending tasks to reduce overhead.