Propose CUDA and CoreML provider support

Provider-only configuration update.

CUDA: FAU A100 execution previously passed.
CoreML: pending macOS testing by Tillmann.
CPU fallback retained.
No model weights or preprocessing changes.

Review PR only; do not merge before macOS validation and release approval.

config.yaml

@@ -48,6 +48,8 @@ runtime:
     - batch
     - 384
   execution_providers:
+  - CUDAExecutionProvider
+  - CoreMLExecutionProvider
   - CPUExecutionProvider
 preparation:
   tokenizer_path: tokenizer