GraphProto loses its sparse tensor initializer going through the optimizer.

Bug Report

Is the issue related to model conversion?

No.

Describe the bug

Start with an ONNX model with a sparse initializer. Run model optimizer with option "eliminate_unused_initializer". Now the returned model no longer has the original sparse initializer.

System information

• OS Platform and Distribution (e.g. Linux Ubuntu 16.04): Linux RedHat5
• ONNX version (e.g. 1.7): 1.6
• Python version: 3.8.2
• GCC/Compiler version (if compiling from source): gcc7
• CMake version:
• Protobuf version: 3
• Visual Studio version (if applicable):

Reproduction instructions

• Describe the code to reproduce the behavior.

import onnx
from onnx import optimizer
model = onnx.load('model.onnx')
model.graph.sparse_initializer
optimized_model = optimizer.optimize(model)
optimized_model.graph.sparse_initializer

• Attach the ONNX model to the issue (where applicable)
  A text version of the ONNX model is attached.

Expected behavior

The sparse initializer should persist in the optimized model.

Notes

When the optimizer converts GraphProto to onnx::Graph in the following stack, the sparse initializer is not carried over:

#0 onnx::graphProtoToGraph (gp=..., nested=nested@entry=false)
at .../onnx/common/ir_pb_converter.cc:188
onnx/onnx#1 0x0000000000568d3f in onnx::ImportModelProto (mp=...)
at .../onnx/common/ir_pb_converter.cc:338
onnx/onnx#2 0x000000000054e39a in onnx::optimization::Optimizer::optimize (
this=this@entry=0x7fffffffcf00, mp_in=...)
at .../onnx/optimizer/optimize.h:26
onnx/onnx#3 0x0000000000534d3c in onnx::optimization::Optimize (mp_in=..., names=...)
at .../onnx/optimizer/optimize.cc:32
rank2-identity.txt

Start at onnx/common/ir_pb_converter.cc:188 and follow the GraphProto-to-Graph conversion used by onnx/optimizer/optimize.h and optimize.cc. Run the supplied Python reproduction with the attached model and verify that the sparse initializer remains in optimized_model.graph.sparse_initializer when elimination is enabled.