[DataLoader] Fix Windows worker exit detection, fix test_proper_exit

[ssnl]

Currently, in test_proper_exit,

1. we do not kill the correct input pid in the kill_pid function
   

   pytorch/test/test_dataloader.py

   Lines 325 to 329 in fe15d6a

   	def kill_pid(pid):
   	if IS_WINDOWS:
   	os.system('taskkill /PID ' + str(os.getpid()) + ' /F')
   	else:
   	os.kill(os.getpid(), signal.SIGKILL)

2. the Windows command that detects process status doesn't actually work
   

   pytorch/test/test_dataloader.py

   Lines 641 to 646 in fe15d6a

   	def _is_process_alive(pid, pname):
   	# There is a chance of a terminated child process's pid being reused by a new unrelated process,
   	# but since we are looping this check very frequently, we will know that the child process dies
   	# before the new unrelated process starts.
   	if IS_WINDOWS:
   	command = 'tasklist | find "{}" /i'.format(pid)

3. worker_error and worker_kill cases (sometimes?) are not tested because the workers may exit naturally due to the pre-fetching mechanism and a too small dataset size / batch size.

In this PR, I, in separate commits:

1. Install psutil (a python package specifically built for process monitoring) on some CI builds. (Linux builds installation are done in Install psutil for dataloader tests pietern/pytorch-dockerfiles#29 install psutil and librosa on non-conda builds pietern/pytorch-dockerfiles#30 Update pytorch docker version to 278 ossci-job-dsl#36 and Bump CircleCI docker version to 278 #15795).

2. Rewrite test_proper_exit with psutil so we

   1. do not rely on the hacky is_process_alive

      pytorch/test/test_dataloader.py

      Lines 640 to 653 in fe15d6a

      	@staticmethod
      	def _is_process_alive(pid, pname):
      	# There is a chance of a terminated child process's pid being reused by a new unrelated process,
      	# but since we are looping this check very frequently, we will know that the child process dies
      	# before the new unrelated process starts.
      	if IS_WINDOWS:
      	command = 'tasklist | find "{}" /i'.format(pid)
      	else:
      	command = 'ps -p {} -o comm='.format(pid)
      	p = subprocess.Popen(command, stdout=subprocess.PIPE, shell=True)
      	(output, err) = p.communicate()
      	p_status = p.wait()
      	output = output.decode('utf-8')
      	return pname in output

   2. increase the #task per worker so worker_error and worker_kill properly trigger
   3. test error message content to ensure that the loader exits with correct message corresponding to each exiting scenario.

3. Fix Windows data loader not having any mechanism to detect worker failures.

[facebook-github-bot]

@soumith is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[ssnl]

@soumith I observed some flakiness so I increased the test join time out. Please land again when you have time. :)

[ssnl]

lint is already broken on master

[facebook-github-bot]

@soumith is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.