fix(packaging): drop Newton/Warp backend to match Isaac-only re-scope (closes #89)

[yinsong1986]

What

Commits the repo to Isaac-Sim-only by removing the Newton/Warp backend
from code + packaging, so they match the docs, README, and umbrella
roadmap (#8). This is Option A from #89 (the likely maintainer intent,
since #8 + the README/docs were deliberately re-scoped to Isaac-only and
all Newton issues — #18–#21, #34, #37–#41 — were closed).

Why

The docs/roadmap said Isaac-only, but the code and packaging still shipped a
full Newton/Warp backend: a reader following the docs would be surprised that
pip install 'strands-robots-sim[newton]' worked and
create_simulation("newton") resolved. The two disagreed; this makes them
agree.

Changes

Packaging (pyproject.toml)

• Removed the [newton] extra (warp-lang / newton-physics).
• Removed newton from the [all] extra; kept [all] as an alias of
  [isaac] so existing pip install 'strands-robots-sim[all]' invocations
  keep working.
• Removed the newton and warp strands_robots.backends entry points.
• Dropped newton/warp from keywords and the "Newton" mention from the
  package description.
• hatch run test now collects only strands_robots_sim/isaac/tests/.

Package layout

• Deleted strands_robots_sim/newton/ (the NewtonSimulation stub +
  tests/).
• strands_robots_sim/__init__.py: dropped the NewtonSimulation mention.
• isaac/{loaders,procedural,config}.py: removed stale comments pointing at
  the deleted newton module.

Docs / examples (Newton mentions + a docs↔packaging gap)

• isaac/tests/test_entrypoint.py + docs/getting-started/installation.md:
  corrected the [isaac] dependency description — it pins
  isaacsim / isaaclab / usd-core, not warp-lang (the docs
  previously claimed warp-lang was pulled in, which was a docs-ships /
  packaging-absent gap).
• tests/README.md, examples/MIGRATION.md, examples/mujoco_gs/README.md,
  examples/libero/libero_backend_matrix.py: dropped Newton rows / examples /
  driver references.
• .github/workflows/test-lint.yml: dropped the Newton CI comment.

Test surface

• hatch run lint — passes (black / isort / flake8 across
  strands_robots_sim + examples).
• hatch run test — 184 passed, 31 skipped (skips are GPU-gated Isaac
  tests). No new failures introduced; the Newton stub-shape suite is removed
  along with the package it pinned.
• import strands_robots_sim smoke — OK.

Acceptance criteria (from #89)

• pyproject.toml (description, extras, entry points, hatch test
  paths), the strands_robots_sim/ package layout, the README/docs, and Umbrella: re-scope robots-sim as the heavy NVIDIA simulation backends for strands-robots #8
  all tell the same Isaac-only story about Newton/Warp scope.
• No "ships in docs but absent in packaging" (or vice-versa) gaps remain
  (also fixed the warp-lang-in-[isaac] doc claim that had no packaging
  backing).

Out of scope (follow-ups)

• The docs reference an isaac_sim entry-point alias
  (docs/api-reference.md, docs/architecture.md, docs/backends/isaac.md,
  docs/getting-started/installation.md, isaac/tests/test_entrypoint.py
  docstring) but pyproject.toml only declares the isaac entry point. That
  is a separate Isaac-aliasing docs↔packaging discrepancy, unrelated to
  Newton/Warp — worth its own issue rather than expanding this PR.
• examples/so101_curobo/* references to warp are about the Warp
  library version conflict with cuRobo (Isaac kit bundles warp 1.5 vs
  cuRobo needing >=1.14), not the Newton backend — intentionally left alone.

Project board

Please move #89 to In review on the
Strands Labs - Robots board
if it isn't moved automatically.

[yinsong1986]

Merged main in and resolved conflicts in docs/getting-started/installation.md and strands_robots_sim/isaac/tests/test_entrypoint.py (both prose/docstring-only conflicts where this PR's Newton/Warp removal and the base's #91 Isaac Sim 4.5.x pin alignment were independent edits — kept the base's more precise 4.5.x wording, which now matches the merged pyproject.toml). pyproject.toml auto-merged cleanly, preserving both the Newton/Warp drop and the 4.5.x pins. Lint + tests green (184 passed, 31 skipped). Ready for re-review.

[cagataycali]

Heads-up: #89 was just closed by #94, which landed the same Option A direction this PR takes (delete strands_robots_sim/newton/, drop the [newton] extra and newton/warp entry points, keep [all] as an [isaac] alias, and scrub the Newton doc/example drift). As a result this PR now shows CONFLICTING and its target issue is already resolved, so it looks superseded.

Before recommending closure I diffed the two change sets and they cover the same surface (pyproject extras/entry-points/keywords/description, strands_robots_sim/newton/ removal, examples/MIGRATION.md, examples/libero/libero_backend_matrix.py, tests/README.md, examples/mujoco_gs/README.md, the installation.md warp-lang doc fix, and the test-lint.yml comment). #94 additionally pins the decision with a TestNewtonRemoved regression suite so it cannot drift back.

The one genuinely-distinct follow-up both PRs flagged as out of scope — the docs claiming an isaac_sim entry-point alias that pyproject.toml never declares — is now tracked separately as #95.

Suggest closing this as superseded by #94 unless there's a delta I missed; happy to be corrected if so.

[yinsong1986]

Merged main in (merge commit 07f49a0) and resolved conflicts in pyproject.toml, strands_robots_sim/init.py, strands_robots_sim/isaac/config.py, strands_robots_sim/isaac/procedural.py, .github/workflows/test-lint.yml, docs/getting-started/installation.md, examples/MIGRATION.md, examples/libero/libero_backend_matrix.py, and examples/mujoco_gs/README.md. All nine conflicts were prose/comment-level wording differences arising because upstream #94 landed the same Newton/Warp removal in parallel; both sides expressed the identical Isaac-only intent, so no behavioural lines diverged. Kept the more complete/specific wording on each side (no information lost). The upstream TestNewtonRemoved contract suite merged cleanly and passes. hatch run lint and hatch run test are green (189 passed, 31 GPU/USD-gated skips, 0 failures). Ready for re-review.

[yinsong1986]

Thanks for the careful diff — your read is correct and I confirmed it independently against current main (post-#94 merge).

What I checked:

• chore: remove Newton/Warp backend (Isaac-only re-scope) #94 is merged (mergedAt 2026-06-17T15:08:18Z) and Newton/Warp still shipped in code + packaging despite the Isaac-only re-scope (#8) #89 is closed by it, so this PR's target issue is already resolved.
• strands_robots_sim/newton/ no longer exists on main (git ls-tree -r upstream/main | grep -i newton → empty), confirming the directory removal already landed.
• Diffing this branch against current main (git diff upstream/main..fix/drop-newton-warp-isaac-only), the only residual changes are prose rewording of the same decision — line re-wrapping and singular/plural ("entry-point plugin(s)", "heavy GPU-only backend(s)") in pyproject.toml, strands_robots_sim/__init__.py, examples/MIGRATION.md, examples/libero/libero_backend_matrix.py, examples/mujoco_gs/README.md, installation.md, strands_robots_sim/isaac/procedural.py, and the test-lint.yml comment. No functional delta: the extras/entry-points/keywords/description edits, the directory removal, and the [all]→[isaac] alias are all already on main via chore: remove Newton/Warp backend (Isaac-only re-scope) #94.

So there's no delta worth a forward-fix here. Agreed this is superseded by #94, with the one genuinely-distinct follow-up (docs claiming an isaac_sim entry-point alias pyproject.toml never declares) already tracked as #95. No objection to closing #92 as superseded.

[cagataycali]

Closing as superseded by #94, which landed the same Isaac-only re-scope (Newton/Warp removal, [newton] extra + entry points dropped, [all] kept as an [isaac] alias, doc/example scrub) and is now on main. Per the discussion above we confirmed there is no functional delta remaining on this branch versus main after #94 — only equivalent prose rewording. The one genuinely-distinct follow-up (docs claiming an isaac_sim entry-point alias pyproject.toml never declares) is tracked separately in #95. Thanks for the parallel work and the careful diff confirming equivalence.