Proxies should never be used when querying GCE metadata server

[thobrla]

Similar changes to https://codereview.appspot.com/187380044/ should be made.

[dhermes]

I filed googleapis/google-cloud-python#757 to make sure we are also doing the right thing.

AFAICT we get around autodetected proxies (and urllib magic overhead) by directly using the httplib library:

from six.moves.http_client import HTTPConnection
import socket

host = '169.254.169.254'
uri_path = '/computeMetadata/v1/project/project-id'
headers = {'Metadata-Flavor': 'Google'}
connection = HTTPConnection(host, timeout=0.1)

RESPONSE_BODY = None
try:
    connection.request('GET', uri_path, headers=headers)
    response = connection.getresponse()
    if response.status == 200:
        RESPONSE_BODY =  response.read()
except socket.error:  # socket.timeout or socket.error(64, 'Host is down')
    pass
finally:
    connection.close()