feat: Fabric-accelerated get/set_local_poses via indexedfabricarray

- get_local_poses / set_local_poses via indexed fabric kernels
- get_scales / set_scales with full Fabric acceleration
- Per-view dirty tracking with automatic re-propagation
- Topology-change recovery via _rebuild_fabric_arrays
- 13 new integration tests (49 total pass)

Author: pv-nvidia

source/isaaclab/test/utils/warp/test_fabric_kernels.py

@@ -10,7 +10,6 @@
 isaaclab_physx/test/sim/test_views_xform_prim_fabric.py.
 """
 
-import pytest
 import warp as wp
 
 wp.init()

source/isaaclab_physx/changelog.d/fabric-local-poses.rst

@@ -0,0 +1,15 @@
+Added
+^^^^^
+
+* Added Fabric-accelerated ``get_local_poses`` / ``set_local_poses`` to
+  :class:`~isaaclab_physx.sim.views.FabricFrameView`.
+
+  Local-pose operations now use ``wp.indexedfabricarray`` to read/write
+  ``omni:fabric:localMatrix`` directly on the GPU, propagating between
+  parent world matrices and child local/world matrices via Warp kernels
+  without round-tripping through USD.
+
+* Added per-view dirty tracking: ``set_local_poses`` marks the world matrix
+  dirty and vice-versa, triggering automatic re-propagation on the next read.
+
+* Added ``_rebuild_fabric_arrays`` topology-change recovery.