Add task-first GenSim action and orchestration pipeline

[skywhite1024]

Note

Superseded by native GitHub Stack #539 (PRs #532 through #538). Review and merge the incremental stack instead of this closed aggregate PR.

Description

This PR introduces a task-first GenSim pipeline that connects scene preparation, symbolic task planning, direct AtomicAction graph execution, and auditable workflow recovery.

The main changes are:

• add Action Engine v2 contracts, capability registration, SeedGraph compilation, planning, runtime execution, recovery, recording, and A/B evaluation;
• add Task Engine orchestration with unified prepare, run, and run-all CLI flows across image, image-edit, scene, and scene-edit inputs;
• extend Scene Engine generation/editing, semantic scene graphs, layout, scale/orientation handling, import/export, and final inspection;
• integrate motion planning, coordinated dual-arm behavior, grasp collision filtering, task-level validation, and deterministic workflow recovery;
• add architecture documentation and broad Action Engine, Task Engine, and Scene Engine regression coverage.

The change consolidates cross-engine ownership under Task Engine while keeping Scene Engine responsible for scene materialization and Action Engine responsible for graph compilation and execution.

Issue: none linked.

Dependencies: no new third-party Python dependency is introduced. Runtime workflows continue to use the existing optional model, image/geometry service, simulator, and motion-planning integrations.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Enhancement (non-breaking change which improves an existing functionality)
• New feature (non-breaking change which adds functionality)
• Breaking change (existing functionality will not work without user modification)
• Documentation update

Validation

• git diff --check origin/main...origin/ljd/action_engine_refactor
• black --check --diff --color ./ — one file requires reformatting: tests/gen_sim/action_engine/config/test_runtime_policy.py; 845 files pass.
• pytest -q tests/gen_sim/action_engine tests/gen_sim/task_engine tests/gen_sim/scene_engine — 703 passed, 3 failed, 9 warnings. The failures are stale expectations for CoordinatedPickment.middle_empty_ratio (0.4 expected, 0.7 configured) and camera recording interval_step (5 expected, 1 configured).
• Full simulator/GPU end-to-end validation was not run locally.

The source branch is currently four commits behind main; it should be synchronized before merge so CI validates the final integration state.

Screenshots

Not included. Generated SeedGraph PNGs and runtime artifacts are covered by the visualization and generation tests.

Checklist

• I have run the black . command to format the code base.
• I have made corresponding changes to the documentation.
• I have added tests that prove my fix is effective or that my feature works.
• Dependencies have been updated, if applicable (no new dependency required).

[greptile-apps]

Too many files changed for review (211 files, 100 file limit).

Bypass the limit by tagging @greptile-apps to review.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of lines (20,000). Try reducing the number of changed lines and requesting a review from Copilot again.