Summary

• add CPU-only ONNX Runtime builds for macOS arm64 and Linux x64
• keep DirectML and CPU providers on Windows x64
• normalize provider selection and platform-specific MLRuntime/native library names
• package one RID per TLX and document the supported platform matrix
• add platform behavior tests

Local verification

• win-x64 plugin build: passed (0 warnings, 0 errors)
• osx-arm64 plugin build: passed (0 warnings, 0 errors)
• linux-x64 plugin build: passed (0 warnings, 0 errors)
• MLRuntime builds for all three RIDs: passed
• unit tests: 64/64 passed

Follow-up

Cross-platform GitHub Actions workflow updates will be added to this branch separately through the GitHub web editor because the current OAuth token cannot modify workflow files.

Summary by CodeRabbit

• New Features
  • Expanded platform support to Windows x64, macOS Apple Silicon, and Linux x64.
  • Automatically selects DirectML on supported Windows systems and CPU elsewhere.
  • Added runtime-specific TLX packaging for each supported platform.
• Bug Fixes
  • Improved native runtime loading to resolve the correct binaries per platform.
  • Avoided selecting unsupported DirectML configurations and tightened provider behavior during load.
• Documentation
  • Updated installation/setup paths, execution-device defaults, and troubleshooting guidance for cross-platform use.
• Tests
  • Added runtime platform/provider coverage to validate supported combinations and provider normalization.