added parsing of cgroups file

[beniwohli]

due to the somewhat underdefined and volatile format of this file,
the "parsing" got pretty spaghetti-code-y :(

This implements #346 and #347

[axw]

Code looks fine overall. Might be a bit neater using regular expressions?

I think it would be ideal if all agents aligned on the approach used for identifying container IDs. If you disagree with my approach, perhaps let's take it back to elastic/apm#21 to discuss?

[graphaelli]

Worth a test to ensure system hostname is used as pod name when kubernetes info is not detected but partially set by environ? Seems like overkill but figured I'd mention it to prove I reviewed this.

@mock.patch.dict(
    "os.environ",
    {
        "KUBERNETES_NAMESPACE": "namespace",
    },
)
def test_docker_kubernetes_system_info_except_hostname_from_environ():
    # initialize agent only after overriding environment
    elasticapm_client = Client()
    # mock docker/kubernetes data here to get consistent behavior if test is run in docker
    with mock.patch("elasticapm.utils.cgroup.get_cgroup_container_metadata") as mock_metadata, mock.patch(
        "socket.gethostname"
    ) as mock_gethostname:
        mock_metadata.return_value = {}
        mock_gethostname.return_value = "foo"
        system_info = elasticapm_client.get_system_info()
    assert "kubernetes" in system_info
    assert system_info["kubernetes"] == {
        "pod": {"name": "foo"},
        "namespace": "namespace",
    }

[beniwohli]

this would have been the ideal chance to use that new fancy review-as-commit thingy :D set you as author of the commit instead. Not that it makes a huge difference after I squash-merge this :D