fix(test): make the Metal self-tests build on MSVC, pass on macOS and fit the sanitizer job

[lusoris]

Summary

This fixes three hosted CI breaks that #1918 introduced when it built the Metal parity tests as self-tests on every host. The changes are tests only; no library code changes.

• Windows MSVC: two tests named a scenario SC_DEFAULT, which winuser.h defines as 0xF160, so the build failed with C2059. The scenarios are now SC_DEFAULTS.
• macOS clang: test_metal_selftest_integer_psnr crashed with SIGSEGV while writing the JSON output to a temporary file and reading apsnr_* back. The test now reads the aggregate straight from the context's feature collector.
• Sanitizers: test_float_moment_sum (ADR-1497) hit its 120 s timeout under the debug ASan + UBSan build. Its timeout is now 600 s. The job did not produce an empty test list; the log only shows the workflow's script text, which contains that message.
• Ubuntu ARM clang / macOS (aarch64): test_metal_selftest_float_moment fails because NEON and SVE2 add the second moment in lane order. fix(simd): add the NEON and SVE2 float_moment lanes in the scalar's order so aarch64 returns the scalar bits past 2^53 units #1923 fixes that, and this PR changes nothing for it.

Type

• test — test-only

Checklist

• Commits follow Conventional Commits (the commit-msg hook enforces this).
• make format && make lint is green locally. The pre-commit hooks pass, and scripts/dev/preflight.sh --stage msvcism passes on the touched files.
• Unit tests pass. I ran the sanitizer job's own invocation locally: the same meson setup flags, the same introspect-and-exclude list, and run_meson_test.py -- -C build --print-errorlogs $TESTS.
  • It enumerated 293 tests, 287 passed and 2 were skipped.
  • All 18 test_metal_selftest_* tests passed, and test_float_moment_sum passed in 35 s.
  • Four DNN tests failed only because this host's ONNX Runtime differs. The hosted job has no ONNX Runtime and does not run them.
• If I touched any SIMD/GPU code path, I ran /cross-backend-diff and the worst ULP is ≤ 2. — Not applicable: only tests changed.
• If I touched a feature extractor with SIMD/GPU twins, I either updated every twin or listed the gap under "Known follow-ups" below.
• If I added a new .c / .cpp / .cu / .h / .hpp, it has the appropriate license header (see CONTRIBUTING.md).
• If this is a breaking change, the commit message uses ! or BREAKING CHANGE: and the migration path is documented below. — Not breaking.
• If this PR adds an ADR, the ADR row lives in docs/adr/_index_fragments/<NNNN-slug>.md and the slug is appended to docs/adr/_index_fragments/_order.txt. — No ADR.

Bug-status hygiene (ADR-0165)

• docs/state.md: T-METAL-SELFTESTS-HOSTED-CI-2026-10-03 is under Recently closed.

Netflix golden-data gate (ADR-0024)

• I did not modify any assertAlmostEqual(...) score in the Netflix golden Python tests.

Deep-dive deliverables (ADR-0108)

• Research digest — no digest needed: test-only CI repair.
• Decision matrix — no alternatives: only-one-way fix (rename the colliding identifier, use the existing test accessor, size the timeout).
• AGENTS.md invariant note — core/test/AGENTS.d/metal-parity-tests.md: no identifiers that windows.h defines, the MinGW check command, and aggregates read through the collector accessor.
• Reproducer / smoke-test command — below.
• CHANGELOG fragment — no changelog: test-only, with no user-visible change.
• Rebase note — no rebase impact: test files only.

Reproducer

# MSVC's C2059, reproduced with MinGW. Run it on the parent commit to see the error, and here to see it gone:
x86_64-w64-mingw32-gcc -fsyntax-only -std=gnu11 -include windows.h -DVMAF_METAL_TWIN_SELFTEST \
  -Icore/include -Icore/src -Icore/test -I<build>/include -I<build>/src core/test/test_metal_integer_motion_parity.c
# The sanitizer job, locally:
cd core && meson setup build-asan -Db_sanitize=address -Dc_args="-fsanitize=undefined -fno-sanitize=function" \
  -Dcpp_args="-fsanitize=undefined -fno-sanitize=function" -Denable_cuda=false -Denable_sycl=false \
  --buildtype=debug -Db_lto=false -Db_lundef=false && meson compile -C build-asan
ASAN_OPTIONS=halt_on_error=1:detect_leaks=1 ./build-asan/test/test_metal_selftest_integer_psnr

Known follow-ups

• The macOS SIGSEGV happened inside the JSON round trip. It does not reproduce on Linux, under ASan or UBSan or without them, so I did not isolate the exact fault. The hosted macOS job on this PR will show whether the new collector read passes there.