Handle name conflict (attribute parameter) when converting back to python

[gramalingam]

Attribute parameters and normal values (like tensors) use the same namespace in onnxscript/python. This can cause a conflict when converting ONNX proto back to onnxscript (since these are different namespaces in ONNX). Examples of where this happens shown in test-case below: eg., when an attribute-parameter "yyy" is used as a value, the onnxscript translator introduces a value called "yyy" which is bound to the attribute-value "yyy".

Fix this in the exporter.

[codecov]

Codecov Report

Attention: 17 lines in your changes are missing coverage. Please review.

Comparison is base (10f9a1f) 78.55% compared to head (632d44b) 78.52%.

Files	Patch %	Lines
onnxscript/backend/onnx_export.py	84.28%	8 Missing and 3 partials ⚠️
onnxscript/backend/onnx_export_test.py	50.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1166      +/-   ##
==========================================
- Coverage   78.55%   78.52%   -0.03%     
==========================================
  Files         118      118              
  Lines       15154    15231      +77     
  Branches     1620     1635      +15     
==========================================
+ Hits        11904    11960      +56     
- Misses       2872     2891      +19     
- Partials      378      380       +2     

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

[github-actions]

Test Results

         18 files  ±0         18 suites  ±0   1h 9m 40s ⏱️ + 11m 35s
  11 188 tests +1    8 343 ✔️ +1      2 792 💤 +1       53 ❌  - 1 
160 011 runs  +8  36 686 ✔️ +9  121 282 💤 ±0  2 043 ❌  - 1 

For more details on these failures, see this check.

Results for commit 632d44b. ± Comparison against base commit 10f9a1f.

This pull request skips 1 test.

onnxscript.tests.function_libs.torch_lib.ops_test.TestOutputConsistencyFullGraphCPU ‑ test_output_match_opinfo__nn_functional_max_pool3d_cpu_float32

♻️ This comment has been updated with latest results.