Migrate onnxscript converter to use onnx ir

[gramalingam]

Migrate onnxscript converter to use onnx ir.

[codecov]

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
11830	1	11829	2253

View the top 1 failed test(s) by shortest run time

onnxscript.rewriter.ort_fusions.rms_normalization_test.TestRmsNormalization::test_smollm

Stack Traces | 8.65s run time

onnxscript\rewriter\ort_fusions\rms_normalization_test.py:19: in test_smollm
    original_outputs = ort_run("original", model, inputs)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxscript\rewriter\ort_fusions\_test_utils.py:19: in ort_run
    model_proto.SerializeToString(), options, providers=providers
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   MemoryError

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

In general, this class of problem is fixed by ensuring that any local variable is always initialized before being read, or by removing/renaming suspicious locals so the analysis is straightforward. Here, the simplest, behavior‑preserving fix is to avoid using the name type as a local variable in value_to_type. That variable is only used as the element type for a sequence (ir.SequenceType(type)), and analogous code elsewhere uses names like elem_type. We can rename the variable to elem_type and update its single use. This does not change logic or outputs, it only clarifies intent and ensures static analysis no longer thinks there is a potentially uninitialized local.

Concretely, in onnxscript/_internal/utils.py within value_to_type, in the elif isinstance(val, list) branch, update the line type, shape = value_to_type(val[0]) to elem_type, shape = value_to_type(val[0]) and the following return ir.SequenceType(type), shape to return ir.SequenceType(elem_type), shape. No imports or additional definitions are required, and no other parts of the file need to change.