ci: speed up wheel builds, defer exotic arches to release tags

- Build aarch64 wheels on native ubuntu-24.04-arm runners instead of
  QEMU emulation (massive speedup on every push/PR).
- Move ppc64le and s390x QEMU builds into a separate job that only
  runs on refs/tags/v* and skips tests (still extremely slow under
  emulation, but no longer gates regular CI).
- Drop EOL CPython 3.8 builds, which also removes the macOS
  x86_64-only-installer warning.
- Include arch in artifact names to make them easier to identify.

.github/workflows/wheels.yml

@@ -15,12 +15,9 @@ jobs:
             arch: x86_64
           - os: ubuntu-latest
             arch: i686
-          - os: ubuntu-latest
+          # Native ARM runner — no QEMU, drastically faster than emulated aarch64
+          - os: ubuntu-24.04-arm
             arch: aarch64
-          - os: ubuntu-latest
-            arch: ppc64le
-          - os: ubuntu-latest
-            arch: s390x
 
           - os: macos-15
             arch: x86_64
@@ -41,8 +38,38 @@ jobs:
     steps:
       - uses: actions/checkout@v5
 
+      - name: Build wheels
+        uses: pypa/cibuildwheel@v3.4.1
+        with:
+          output-dir: wheelhouse
+        env:
+          CIBW_ARCHS: ${{ matrix.arch }}
+          # Drop EOL CPython 3.8 (also avoids the macOS x86_64-only-installer warning)
+          CIBW_SKIP: "cp38-*"
+          CIBW_TEST_REQUIRES: pytest pytest-asyncio
+          CIBW_TEST_COMMAND: pytest {project}
+          # Only needed to make the macosx arm64 build work
+          CMAKE_OSX_ARCHITECTURES: "${{ matrix.arch == 'arm64' && 'arm64' || '' }}"
+
+      - uses: actions/upload-artifact@v6
+        with:
+          name: cibw-wheels-${{ matrix.os }}-${{ matrix.arch }}-${{ strategy.job-index }}
+          path: ./wheelhouse/*.whl
+
+  # Exotic arches built via QEMU emulation — very slow (often >1h), so only run
+  # on release tags rather than every push/PR.
+  build_wheels_exotic:
+    name: Build wheels (exotic) ${{ matrix.arch }}
+    if: startsWith(github.ref, 'refs/tags/v')
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [ppc64le, s390x]
+    steps:
+      - uses: actions/checkout@v5
+
       - name: Set up QEMU
-        if: runner.os == 'Linux'
         uses: docker/setup-qemu-action@v4
 
       - name: Build wheels
@@ -51,17 +78,13 @@ jobs:
           output-dir: wheelhouse
         env:
           CIBW_ARCHS: ${{ matrix.arch }}
-          CIBW_TEST_REQUIRES: pytest pytest-asyncio
+          CIBW_SKIP: "cp38-*"
-          CIBW_TEST_COMMAND: pytest {project}
+          # Tests under QEMU are extremely slow; skip them for these arches.
-          # cibuildwheel can build cp38 macOS universal2/arm64 wheels but can't test the
+          CIBW_TEST_SKIP: "*"
-          # arm64 slice because the cp38 installer is x86_64-only. Silence the warning.
-          CIBW_TEST_SKIP: "cp38-macosx_*:arm64"
-          # Only needed to make the macosx arm64 build work
-          CMAKE_OSX_ARCHITECTURES: "${{ matrix.arch == 'arm64' && 'arm64' || '' }}"
 
       - uses: actions/upload-artifact@v6
         with:
-          name: cibw-wheels-${{ matrix.os }}-${{ strategy.job-index }}
+          name: cibw-wheels-${{ matrix.arch }}
           path: ./wheelhouse/*.whl
 
   build_sdist: