Added GraspCandidateBatch to preserve multiple grasp poses

[matafela]

Description

• Added GraspCandidateBatch to preserve multiple grasp poses, costs, opening widths, validity masks, and stable IDs.
• Added candidate-aware atomic planning with optional same-arm motion prefixes followed by PickUp: transit → approach → close/settle → lift.
• Added example: python scripts/tutorials/atomic_action/pickup.py --n_affordance_multi_gen 8
• Document: docs/design/multi_grasp_atomic_trajectory_generation_design.md
• Added SceneReplicaPool to schedule logical candidates across verified physical scene replicas.
• Added per-candidate filtering for invalid grasps, IK/FK failures, joint limits, and trajectory discontinuities, with bounded rejection diagnostics.
• Exported compact (B, N, D) trajectory batches with timing, valid lengths, padding masks, and provenance; supports partial and empty results.
• Fixed nonstandard URDF link-origin handling and example resource cleanup.
• Document:
• Validated in embodichain2: GPU smoke test produced (8, 260, 8) across four environments; 1,261 CPU tests passed. One existing Python 3.10 notes compatibility test remains failing.
• Scope: planning-only output; physical grasp validation, held-object suffixes, and expert-dataset persistence remain outside this implementation.

Type of change

• Enhancement (non-breaking change which improves an existing functionality)
• New feature (non-breaking change which adds functionality)

Checklist

• I have run the black . command to format the code base.
• I have made corresponding changes to the documentation
• Public API changes are reflected in the API docs (python docs/scripts/check_api_docs.py), if applicable
• I have added tests that prove my fix is effective or that my feature works
• Dependencies have been updated, if applicable.