Added converter for torch.Tensor.expand_as() method

[maronuu]

Description

This PR is related to Issue #565, but this issue cannot be closed yet even when this PR is merged.

A converter for torch.Tensor.expand(size) already exists, but one for torch.Tensor.expand_as(other) did not.
Users can convert expand_as by replacing all expand_as(other) with expand(other.size()), but it should not be a favorable way.

I made a converter torch2trt/converters/expand_as.py similar to torch2trt/converters/expand.py, with some new test cases.

The result of testing in my environment is as follows:

Environment

Ubuntu 20.04
Python 3.8.8
torch 1.8.1
torch2trt 0.3.0
cuda 11.1
cudnn 8.1.1
TensorRT 7.2.2

Test Result

| torch2trt.converters.expand_as.test_tensor_expand_as_scalar | float32 | [1] | {} | 0.00E+00 | nan | 0.00E+00 | 3.4e+05 | 1.88e+04 | 0.014 | 0.0646 |
| torch2trt.converters.expand_as.test_tensor_expand_as_singledim | float32 | [(1, 1, 3, 3)] | {} | 0.00E+00 | nan | 0.00E+00 | 3.3e+05 | 1.97e+04 | 0.0129 | 0.0631 |
| torch2trt.converters.expand_as.test_tensor_expand_as_multidim | float32 | [(1, 1, 1, 3)] | {} | 0.00E+00 | nan | 0.00E+00 | 3.2e+05 | 1.78e+04 | 0.0138 | 0.0635 |
| torch2trt.converters.expand_as.test_tensor_expand_as_singledim_half | float16 | [(1, 1, 3, 3)] | {} | 0.00E+00 | nan | 0.00E+00 | 3.17e+05 | 1.52e+04 | 0.0167 | 0.0877 |
| torch2trt.converters.expand_as.test_tensor_expand_as_multidim_half | float16 | [(1, 1, 1, 3)] | {} | 0.00E+00 | nan | 0.00E+00 | 3.25e+05 | 1.53e+04 | 0.0134 | 0.0784 |
NUM_TESTS: 5
NUM_SUCCESSFUL_CONVERSION: 5
NUM_FAILED_CONVERSION: 0
NUM_ABOVE_TOLERANCE: 0
NUM_pSNR_TOLERANCE: 0

[github2016-yuan]

@maronuu Have you converted yolov5s.pt to yolov5s_trt.pt yet?

[maronuu]

@github2016-yuan
No, I don't work on YOLOv5.
If you are looking for the YOLOv5 implementation in TensorRT, this repository might be helpful.