fix: issues related to TensorRT-Model-Optimizer

[inisis]

Related to NVIDIA/TensorRT-Model-Optimizer#478

1. keep original shape info when shape inference failed
2. fix resize output calculation, using round instead of floor
3. fix precision issue with conv+bn merge

[codecov]

Codecov Report

❌ Patch coverage is 96.77419% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.52%. Comparing base (c6dbd11) to head (2e2f74c).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
tests/test_nvidia.py	96.62%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #221      +/-   ##
==========================================
+ Coverage   90.37%   90.52%   +0.15%     
==========================================
  Files          41       42       +1     
  Lines        3615     3704      +89     
==========================================
+ Hits         3267     3353      +86     
- Misses        348      351       +3     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[inisis]

There is a contradiction here, in onnx operator doc, it uses floor, where in ort, it uses round

https://onnx.ai/onnx/operators/onnx__Resize.html#summary

https://github.com/microsoft/onnxruntime/blob/977efe4788b2ee24371523b5fa14dd02efcd4942/onnxruntime/core/providers/cpu/tensor/upsample.cc#L70