[DD001] Generalize the parity test so it works for all four demos, not just demo1

[slarson]

• Roadmap Phase: Phase 0
• Required Capabilities: python, physics
• Audit note: tests/test_demo1_backend_parity.py (256 lines) is already a fully-functional parity harness — parses position buffers with .times.txt sidecar handling, computes per-particle L2 + cube-center/extent drifts + init/mean/final L2, has explicit THRESHOLDS dict and 5 pass/fail gates, JSON + human-readable output. The proposed new file scripts/backend_parity_test.py would duplicate ~80% of this. Milestone: Logically validation infrastructure, but co-ships with v0.0.9 (Phase 0 Modernization) because the per-demo parity issues [DD001] Run the demo1 (cube drop) parity test automatically in CI on every PR #234/[DD001] Add a parity check for demo2 (membrane permeability) — depends on #236 #235/[DD001] Add a parity check for worm_alone (passive worm body) — depends on #236 #236/[DD001] Turn on muscle drive in the worm_swim test and add a parity check (depends on #236, #241) #239 all depend on the generalized harness — splitting these across milestones would force v0.0.9 to ship without working CI gates.
• DD Section to Read: DD001 §Cross-Backend Parity Requirements, §Backend Graduation Criteria
• Depends On: None (the per-demo parity issues [DD001] Run the demo1 (cube drop) parity test automatically in CI on every PR #234-[DD001] Add a parity check for worm_alone (passive worm body) — depends on #236 #236, [DD001] Turn on muscle drive in the worm_swim test and add a parity check (depends on #236, #241) #239 depend on this)
• Existing Code to Reuse / Refactor:
  • tests/test_demo1_backend_parity.py — rename to tests/test_backend_parity.py, parameterize on scenario
  • scripts/cross_backend_regression.py — driver that runs N backends; already pluggable via --binary NAME=PATH
  • src/metal_diff/dump_metal_trajectory.py — Metal trajectory dumper writing OpenCL-compatible format
• Approach: Refactor — promote the existing demo1-specific harness into a scenario registry. Per-demo configs (THRESHOLDS, parsers, baselines) live in tests/parity_scenarios/{demo1,demo2,worm_alone,worm_swim}.py. Add --scenario and --substrate CLI flags. Keep the L2 engine and threshold logic that already exists.
• Files to Modify:
  • tests/test_backend_parity.py (rename from test_demo1_backend_parity.py)
  • tests/parity_scenarios/{demo1,demo2,worm_alone,worm_swim}.py (new — per-demo configs)
  • tests/baseline/*_opencl.json (new — committed reference metrics)
• Acceptance Criteria:
  • Runs --scenario demo1 --substrate metal-native and reproduces the existing demo1 parity check
  • Adds entry points for demo2, worm_alone, worm_swim (used by [DD001] Add a parity check for demo2 (membrane permeability) — depends on #236 #235, [DD001] Add a parity check for worm_alone (passive worm body) — depends on #236 #236, [DD001] Turn on muscle drive in the worm_swim test and add a parity check (depends on #236, #241) #239)
  • --substrate cuda-native returns "not implemented yet" error message
  • Emits JSON + human-readable table; exit code reflects pass/fail
  • cross_backend_regression.py continues to work
• Sponsor Summary Hint: The parity engine exists. This issue generalizes the existing demo1-specific harness so demo2/worm_alone/worm_swim each get a parity gate, without rewriting the L2/sampling/threshold core.

---

This issue is part of milestone v0.0.9 — Phase 0 Modernization, derived from DD001 Body Physics Engine Architecture. Related issues in this milestone are filed in the same batch.