feat: add manylinux self-hosted build job and enable version matrix

- Add manylinux self-hosted build workflow job in build.yml
- Configure manylinux matrix with flash-attn versions 2.6.3, 2.7.4, 2.8.3
- Enable Python versions 3.10, 3.11, 3.12 for manylinux builds
- Configure CUDA versions 12.8 and 13.0 for manylinux builds
- Add manylinux_self_hosted to build and release workflow dependencies
- Update create_matrix.py to enable manylinux_self_hosted matrix configuration
- Disable Windows and Windows CodeBuild matrices temporarily

create_matrix.py

@@ -91,19 +91,19 @@ LINUX_ARM64_MATRIX = {
 LINUX_SELF_HOSTED_MATRIX = {
     "flash-attn-version": [
         "2.6.3",
-        # "2.7.4",
-        # "2.8.3",
+        "2.7.4",
+        "2.8.3",
     ],
     "python-version": [
-        # "3.10",
-        # "3.11",
-        # "3.12",
+        "3.10",
+        "3.11",
+        "3.12",
         # "3.13",
-        "3.14",
+        # "3.14",
     ],
     "torch-version": ["2.9.1"],
     "cuda-version": [
-        # "12.8",
+        "12.8",
         "13.0",
     ],
 }
@@ -200,11 +200,14 @@ def main():
                 "linux_self_hosted": False,
                 # "linux_self_hosted": LINUX_SELF_HOSTED_MATRIX,
                 #
+                # "manylinux_self_hosted": False,
+                "manylinux_self_hosted": LINUX_SELF_HOSTED_MATRIX,
+                #
                 "linux_arm64_self_hosted": False,
                 # "linux_arm64_self_hosted": LINUX_ARM64_SELF_HOSTED_MATRIX,
                 #
-                # "windows": False,
-                "windows": WINDOWS_MATRIX,
+                "windows": False,
+                # "windows": WINDOWS_MATRIX,
                 #
                 "windows_code_build": False,
                 # "windows_code_build": WINDOWS_CODEBUILD_MATRIX,