fix: guard optimize/quantize/compile against EPContext models (#256 #257)

[DingmaomaoBJTU]

Summary

winml optimize and winml quantize crash with an ORT NOT_IMPLEMENTED error when given a compiled QNN EPContext model. The crash happens deep inside ORT's session initialization, which gives no indication that the real problem is that the input model is already compiled.

Add an is_compiled_onnx() guard at the start of all three commands that operate on uncompiled ONNX models. Also adds the guard to winml compile to catch re-compilation attempts.

Changes

• src/winml/modelkit/commands/optimize.py — EPContext guard before the optimizer runs
• src/winml/modelkit/commands/quantize.py — EPContext guard before quantization runs
• src/winml/modelkit/commands/compile.py — EPContext guard to prevent re-compilation (bonus, found during review)

All guards use the existing is_compiled_onnx() helper from winml.modelkit.onnx.

Before / After

winml optimize — Before:

Loading model...
Running optimizer...
ERROR ✗ ort_graph failed: ONNX Runtime optimization failed: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED :
  Could not find an implementation for EPContext(1) node with name
  'QNNExecutionProvider_QNN_16222565441986465013_1_0'
Error: Optimization failed: ONNX Runtime optimization failed: ...

winml optimize — After:

Error: model.onnx is a compiled EPContext model and cannot be optimized.
Run 'winml optimize' on the original ONNX model before compilation.

---

winml quantize — Before:

Running quantization...
ERROR: Quantization failed
onnxruntime.capi.onnxruntime_pybind11_state.NotImplemented: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED :
  Could not find an implementation for EPContext(1) node with name
  'QNNExecutionProvider_QNN_16222565441986465013_1_0'
Error: Quantization failed

winml quantize — After:

Error: model.onnx is a compiled EPContext model and cannot be quantized.
Run 'winml quantize' on the original ONNX model before compilation.

---

winml compile (bonus) — After:

Error: model_ctx.onnx is already a compiled EPContext model and cannot be re-compiled.
Run 'winml compile' on the original ONNX model.

Closes #256
Closes #257

[timenick]

Two suggestions on guard placement — see inline comments.

[DingmaomaoBJTU]

Fixed in 90a8503 — moved both guards to fail-fast positions:

• optimize.py: guard now fires right after the model is None check, before any config parsing or console output
• quantize.py: guard now fires right after configure_logging(), before type resolution, output-path computation, or any console output