Goal

Close out the test-parity acceptance criteria of #2123: same-named Isaac tests mirroring the MuJoCo motion-primitive suite, plus a real-Isaac integration test.

Depends on #2155 - the Isaac move_to implementation (and transitively #2154) - if that PR hasn't merged yet, post a comment on this issue and stop.

Scope

1. Parity unit suite: tests/simulation/isaac/test_motion_primitives.py mirroring tests/simulation/mujoco/test_motion_primitives.py case-for-case where the behavior is backend-neutral (validation rejections, gripper metadata semantics, abort reasons, structured error shapes). GPU-dependent cases are skipped in unit CI using the same skip pattern as the existing tests/simulation/isaac/ files. Where a MuJoCo case pins a MuJoCo-only mechanism (e.g. data.ctrl interplay, _SUBSTEPS_PER_TICK), note the omission in a comment rather than porting it emptily - test behavior, not implementation.
2. Integration test (AGENTS.md rule 8): tests_integ/simulation/test_isaac_motion_primitives.py (or matching the existing tests_integ/ layout) running against real Isaac Sim:
   • move_to a named/known-reachable pose; assert convergence within tol.
   • move_to an unreachable target (e.g. beyond the arm's reach but inside the workspace sanity radius); assert the structured error carries the correct abort reason / IK residual.
   • set_gripper open -> closed round trip asserting the gripper joint reaches the registry-mapped range ends.
   • Caution on reset between cases: bug(sim/isaac): world.reset() kills articulation handles on pip Isaac Sim 6.0.x - get_observation returns {} after any reset (+ USD-mangled numeric URDF joint names) #1895 (reset() kills articulation handles) is not a blocker but tests that reset between cases must account for it - prefer one sim session per test or re-acquire handles, and reference tests/simulation/isaac/test_reset_revives_articulations.py for the current state of that behavior.
3. If any of the three primitives are surfaced through agent tools during this work, every exposed parameter needs its own Args: entry (AGENTS.md rule 13, pinned by tests/tools/test_agent_tool_parameter_descriptions.py). If they are not tool-surfaced, this item is a no-op - do not add tool wrappers here.

Constraints

• No production-code changes expected; if a parity test exposes a real behavioral divergence, file it as its own issue and skip/xfail with a reference rather than patching production code in this test-only PR.
• MuJoCo suite untouched and green; hatch run format && hatch run lint && hatch run test green (integration tests run via hatch run test-integ on GPU hardware and are not part of unit CI).
• Changelog fragment changelog.d/<pr-number>-isaac-primitive-parity-tests.md.

Part of #2123. Newton has the same gap; it stays out of scope here (noted follow-up on the parent).

Parent: #2123