feat(gate): Pyodide differential parity gate (Phase 2)

[luisleo526]

Summary

Phase 2 of the Pyodide-target conformance flow (spec: docs/superpowers/specs/2026-06-16-pyodide-conformance-flow-design.md, §4/§6). Adds a CI gate that runs the transpiler inside real wasm32 Pyodide and proves it behaves identically to native CPython on both branches:

• success ⇒ success, byte-identical C++
• failure ⇒ the same failure — same verdict, error, diagnostics, and unexpected-exception type/message

over a 275-fixture corpus (260 valid strategies + 15 failure families). Locally: gate:full → PARITY OK over 275; canary selftest 8/8 in ~0.1s.

Pieces

• PYODIDE_TARGET (314.0.0) — single source of truth; Python/Emscripten derived from Pyodide's lock at gate time.
• First Node tooling in the repo (package.json: pyodide + tar).
• gate/glue.py (canonical glue, runtime-equivalent to the app worker), gate/oracle.py (native, one process, captures unexpected exceptions), gate/compare.mjs (pure comparator), gate/run-gate.mjs (pack → Pyodide → compare → release.json), gate/selftest.mjs (canary).
• tests/gate-corpus/{ok,err}/ corpus.
• .github/workflows/gate.yml — dispatch + nightly + gate-file PRs. PYTHONHASHSEED=0; pinned action SHAs.

Deliberate deferrals (Phase 3)

• Release-blocking wiring (gate as a needs: of release) — Phase 3 with npm delivery. The gate generates release.json here; it does not block tags yet.
• Glue dedup (ship one glue from the npm package).

Test plan

• gate.yml runs green on this PR (clean runner, native Python 3.14 oracle vs wasm32 Pyodide).
• Local: gate:full PARITY OK over 275; gate:selftest 8/8; existing pytest suite unaffected.

🤖 Generated with Claude Code