Fix world_id/env_id conflation in NewtonSiteFrameView for the flat builder

[pascal-roth]

Description

NewtonSiteFrameView._initialize_from_specs resolves world-attached (body-less)
sites by indexing NewtonManager._world_xforms directly with a per-environment
env_id. The flat (non-replicated, replicate_physics=False) builder has
exactly one shared Newton world regardless of environment count, so
_world_xforms has length 1 there — any env_id greater than 0 raises
IndexError: list index out of range.

This is hit whenever a per-environment site with no rigid-body/articulation-root
ancestor is resolved under the flat builder — for example a streaming/follow
camera (streaming_cam_target_prim_path) whose target pattern falls back to the
per-world site path because the target prim itself isn't a rigid body.

File ".../isaaclab_newton/sim/views/newton_site_frame_view.py", line 431, in _initialize_from_specs
    world_xform = NewtonManager._world_xforms[world_id]
                  ~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^
IndexError: list index out of range

The fix falls back to the single shared world when _world_xforms has only one
entry, matching the same "flat builder has fewer per-thing entries than
per-env" pattern already fixed for site injection (#7672) and articulation
bindings (#7673).

Fixes # (no tracked issue — found while wiring up a Newton follow-camera for a
downstream policy-recording script)

Type of change

• Bug fix (non-breaking change which fixes an issue)

Release backport

• Backport this pull request to the active release branch after it merges into develop

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks (ruff check / ruff format)
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective (verified failing without the fix, passing with it)
• I have added a changelog fragment under source/isaaclab_newton/changelog.d/
• I have added my name to CONTRIBUTORS.md (already present)

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; the fallback is limited to the single-world configuration and the original failure path is covered by a regression test.

Summary

• Avoids out-of-bounds _world_xforms access for environment IDs greater than zero.
• Preserves existing indexing for replicated, multi-world builders.
• Adds a multi-environment regression test covering the original failure path.
• Documents the fix in a changelog fragment.

[isaaclab-review-bot]

Isaac Lab Review Bot

The single-world fallback prevents the flat-builder IndexError, but it collapses per-environment world-attached frames onto the same transform, producing incorrect world poses when environment origins differ.

• Design and architecture: Handling the flat builder inside NewtonSiteFrameView is appropriately localized, but mapping every environment to _world_xforms[0] does not preserve the replicated path’s environment-to-world transform semantics. The flat path must retain each environment’s origin when resolving world-attached sites.
• API: No public signatures or result shapes change, and the changelog fragment is present. However, FrameView.get_world_poses() can now return identical, incorrect world-space poses for distinct environments, so the behavioral contract is not correctly restored.
• Implementation: At line 450, all env_ids use the same _world_xforms[0] when only one Newton world exists. Because the same site-local xform is then composed for each entry, environment-root offsets are lost. The regression test reinforces this collapse by expecting identical positions despite configuring multiple environments with nonzero spacing; the implementation and test should instead account for each environment’s origin transform.

Significant concerns. Posted 1 actionable finding inline.

Automated review; human maintainers own approval decisions.

[isaaclab-review-bot]

🟡 Warning · Implementation — Flat-builder fallback collapses per-env world poses

In the replicated path xform is env-relative and world_xform supplies the environment offset. With num_worlds == 1, every world_id in env_ids now composes the same _world_xforms[0] with the same xform, so all environments report an identical world pose and env-root offsets are silently lost (e.g. a follow camera tracking env 0's location for every env). Compose each environment's origin transform in the flat case instead of reusing world 0.