Deprecate Conv2d, ConvTranspose2d and BatchNorm2d

[fmassa]

PyTorch now natively supports empty batches for Conv2d, ConvTranspose2d and BatchNorm2d, so no need to have our own versions anymore.

Follow-up from #2215 which removed those ops. I'm adding the functions back to avoid breaking BC, but I'm raising a error instead to encourage users to replace those ops in their code.

[fmassa]

@pmeier I've started seeing python_type_check errors recently, do you have an idea of what this might be?

[pmeier]

@pmeier I've started seeing python_type_check errors recently, do you have an idea of what this might be?

I'll investigate and get back to you.

---

The problem arises if version.py does not exist in torchvision/. It should be created while running setup.py. Since we install torchvision during the CI run

vision/.circleci/config.yml.in

Lines 131 to 135 in cb65590

	command: |
	pip install --user --progress-bar off numpy mypy
	pip install --user --progress-bar off --pre torch -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
	pip install --user --progress-bar off .
	mypy --config-file mypy.ini

and the creation of the version.py file is unconditional

vision/setup.py

Lines 51 to 61 in cb65590

	def write_version_file():
	version_path = os.path.join(cwd, 'torchvision', 'version.py')
	with open(version_path, 'w') as f:
	f.write("__version__ = '{}'\n".format(version))
	f.write("git_version = {}\n".format(repr(sha)))
	f.write("from torchvision.extension import _check_cuda_version\n")
	f.write("if _check_cuda_version() > 0:\n")
	f.write(" cuda = _check_cuda_version()\n")
	write_version_file()

I am a little puzzled about what is going on. Unfortunately, I don't have to capacity to debug the CI for now. Could you do that? I would start by checking if the file is created where it is supposed to be. Maybe cwd is incorrectly set in setup.py?

[fmassa]

@pmeier we should note that the CI job for python_type_check doesn't compile nor install torchvision, so setup.py is not invoked, so this error is kind of expected I think, following your explanation.

What puzzles me is that this CI was working before and just recently stopped working, and I don't have any ideas why that would be the case.

[pmeier]

CI job for python_type_check doesn't compile nor install torchvision, so setup.py is not invoked,

Isn't L134 below doing exactly that?

vision/.circleci/config.yml.in

Lines 131 to 135 in cb65590

	command: |
	pip install --user --progress-bar off numpy mypy
	pip install --user --progress-bar off --pre torch -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
	pip install --user --progress-bar off .
	mypy --config-file mypy.ini

[fmassa]

Oh, you are right, I missed that line. Let me have a closer look at the commit history

[fmassa]

Let's see if #2246 fixes it (I'm not sure)