Use new GraphInitializers tensor APIs to simplify codebase

[justinchuby]

Accesses to initializer tensors across the test suite still used the old three-layer pattern (initializers["name"].const_value, for value in .values(): value.const_value) introduced before the get_tensor() / tensors() / tensor_items() APIs existed.

Changes

• initializers["name"].const_value → initializers.get_tensor("name") — applied across external_data_test.py, serde_test.py, _safetensors_test.py, _io_test.py, shape_inference_test.py, initializer_deduplication_test.py, constant_manipulation_test.py
• for value in .values(): value.const_value → for tensor in .tensors() — where only the tensor is needed
• for name, value in .items(): value.const_value → for name, tensor in .tensor_items() — where name + tensor are both needed

# Before
external_tensor = model.graph.initializers["tensor1"].const_value
for value in model.graph.initializers.values():
    self.assertIsInstance(value.const_value, ir.Tensor)

# After
external_tensor = model.graph.initializers.get_tensor("tensor1")
for tensor in model.graph.initializers.tensors():
    self.assertIsInstance(tensor, ir.Tensor)

Production code is left unchanged: those loops either require the Value object for operations unrelated to the tensor (naming, replace_all_uses_with, etc.) or deliberately skip None const_values — behaviour that differs from tensors()'s strict-raise semantics.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.57%. Comparing base (4ee65a2) to head (a14f6b7).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #423   +/-   ##
=======================================
  Coverage   84.57%   84.57%           
=======================================
  Files          52       52           
  Lines        6501     6501           
  Branches     1329     1329           
=======================================
  Hits         5498     5498           
  Misses        643      643           
  Partials      360      360           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Copilot]

Pull request overview

Updates the test suite to use the newer GraphInitializers tensor-centric APIs (get_tensor(), tensors(), tensor_items()) instead of repeatedly dereferencing initializer Value.const_value, improving readability and aligning tests with the current public API surface.

Changes:

• Replaced initializers["name"].const_value with initializers.get_tensor("name") in multiple tests.
• Replaced loops over .values()/.items() + .const_value with .tensors() / .tensor_items() where appropriate.
• Kept behavior-sensitive cases using get_tensor() (which returns default when const_value is None) to preserve existing test intent.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/onnx_ir/serde_test.py	Uses initializers.get_tensor() for initializer tensor assertions in serde tests.
src/onnx_ir/passes/common/shape_inference_test.py	Updates initializer access patterns and preserves None-const-value coverage via get_tensor().
src/onnx_ir/passes/common/initializer_deduplication_test.py	Switches initializer tensor reads to get_tensor() in deduplication tests.
src/onnx_ir/passes/common/constant_manipulation_test.py	Updates lifted-initializer verification to compare directly against get_tensor() results.
src/onnx_ir/external_data_test.py	Converts external-data tests to use get_tensor() for initializer tensor access.
src/onnx_ir/_safetensors/_safetensors_test.py	Replaces initializer iteration/access with tensors() and tensor_items() for safetensors tests.
src/onnx_ir/_io_test.py	Uses get_tensor() for initializer tensor assertions in IO tests.

[justinchuby]

These are all tests. So I don't know how useful the API is, especially when the setter is not mirroring it. @gramalingam do you have thoughts?