Add Wheels Build manylinux2014_x86_64 workflow (triton-lang#5619)

In order to execute this workflow with ``workflow_dispatch`` event this
needs to be part of main

Author: atalman

.github/workflows/wheels_v2.yml

@@ -0,0 +1,78 @@
+name: Wheels Build manylinux2014_x86_64
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "20 2 * * *"
+
+jobs:
+
+  Build-Wheels:
+    timeout-minutes: 60
+
+    runs-on: [self-hosted, CPU]
+    permissions:
+      id-token: write
+      contents: read
+
+    steps:
+
+      - name: Prune stale docker containers
+        run: |
+          # If cibuildwheel crashes (or, say, is OOM-killed), it leaves behind a
+          # docker container.  Eventually these consume all the disk space on
+          # this machine.
+          docker container prune -f
+
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      # The LATEST_DATE here should be kept in sync with the one in Patch setup.py
+      - id: check-version
+        name: Check latest version
+        run: |
+          export PACKAGE_DATE=$(python3 -m pip install --user --index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/Triton-Nightly/pypi/simple/ --dry-run triton-nightly== |& grep -oP '(?<=, )[0-9\.]+dev[0-9]+(?=\))' | grep -oP '(?<=dev)[0-9]+')
+          export LATEST_DATE=$(TZ=UTC0 git show --quiet --date='format-local:%Y%m%d%H%M%S' --format="%cd")
+          if cmp -s <(echo $PACKAGE_DATE) <(echo $LATEST_DATE); then
+            echo "new_commit=false" >> "$GITHUB_OUTPUT"
+          else
+            echo "new_commit=true" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Patch setup.py
+        if: ${{ steps.check-version.outputs.new_commit == 'true' }}
+        run: |
+          echo "" >> python/setup.cfg
+          echo "[build_ext]" >> python/setup.cfg
+          echo "base-dir=/project" >> python/setup.cfg
+
+      - name: Build wheels
+        if: ${{ steps.check-version.outputs.new_commit == 'true' }}
+        run: |
+          python3 -m  pip install cibuildwheel --upgrade --user
+          export LATEST_DATE=$(TZ=UTC0 git show --quiet --date='format-local:%Y%m%d%H%M%S' --format="%cd")
+          # Pass MAX_JOBS=4 because, at time of writing, the VM "only" has 32GB
+          # of RAM and OOMs while building if we give it the default number of
+          # workers (2 * NUM_CPUs).
+          #
+          # Sadly, I couldn't make TRITON_BUILD_WITH_CLANG_LLD=1 work.  The
+          # manylinux image has a relatively recent gcc (v10, released 2020),
+          # but its clang is ancient, v3.4, released in 2014 (!).  I tried
+          # installing the prebuilt clang 10 binary distributed by LLVM, and I
+          # quickly ran into Linux DLL hell.  I give up, for now.  Perhaps
+          # manylinux_x_y will save us; I didn't try.
+          export CIBW_ENVIRONMENT="MAX_JOBS=4 TRITON_WHEEL_NAME=triton"
+          export CIBW_MANYLINUX_X86_64_IMAGE="quay.io/pypa/manylinux2014_x86_64:latest"
+          #export CIBW_MANYLINUX_PYPY_X86_64_IMAGE="quay.io/pypa/manylinux2014_x86_64:latest"
+          export CIBW_BEFORE_BUILD="pip install cmake;"
+          export CIBW_SKIP="cp{35,36,37,38}-*"
+          export CIBW_BUILD="cp3{9,10,11,12,13}-manylinux_x86_64"
+          python3 -m cibuildwheel python --output-dir wheelhouse
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: cibw-wheels-manylinux2014-wheels-upload
+          path: ./wheelhouse/*.whl
+
+      - name: Upload wheels to PyPI
+        run: |
+          python3 -m twine upload wheelhouse/* -u __token__ -p ${{ secrets.PYPY_API_TOKEN }}