[Feature] Support for 1D Convolution

[antimora]

I was trying to import my ONNX model but it has failed with the following error:

$ /usr/bin/python3 /workspace/trouble_onnx.py
Traceback (most recent call last):

  File "/workspace/trouble_onnx.py", line 15, in <module>
    mod, params = relay.frontend.from_onnx(onnx_model, shape_dict)

  File "/workspace/python/tvm/relay/frontend/onnx.py", line 1553, in from_onnx
    mod, params = g.from_onnx(graph, opset)

  File "/workspace/python/tvm/relay/frontend/onnx.py", line 1381, in from_onnx
    op = self._convert_operator(op_name, inputs, attr, opset)

  File "/workspace/python/tvm/relay/frontend/onnx.py", line 1481, in _convert_operator
    sym = convert_map[op_name](inputs, attrs, self._params)

  File "/workspace/python/tvm/relay/frontend/onnx.py", line 246, in _impl_v1
    custom_check=dimension_constraint())(inputs[:2], attr, params)

  File "/workspace/python/tvm/relay/frontend/common.py", line 377, in __call__
    raise RuntimeError("Check failed: {}".format(msg))

RuntimeError: Check failed: Only 2d kernel supported.

Code is the latest version (master@1900041447cc7fd4f23a857c4839acbf74cd890c) as of now.

I noticed @tqchen closed this related issue #3518 recently but not explanation was given. Could it be because this related PR was merged?

I would really love if TVM adds 1D Conv support. It would be tremendous for speech processing projects.

To help with debugging and development, I am attaching onnx model and graph image.
trouble.zip

You can run the following python code to reproduce an error:

import onnx
import numpy as np
import tvm.relay as relay

onnx_model = onnx.load('trouble.onnx')

input1 = np.random.rand(1, 192, 100)
input2 = np.random.rand(1)

shape_dict = {'1': input1.shape, 'x.1': input2.shape}
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict)

[masahi]

Conv 1D was added in #4639