Override glog's default for the -logtostderr flag

[klingerf]

In #524 we started setting the -logtostderr=true flag on all of the controller components in the kubernetes configs that are output by the conduit install command. This fixed the issue described in #522, but it would be better to fix this in a way that didn't require modifying every command line invocation of every controller component.

We should be able to call flag.Set in our code to accomplish this, per: https://twitter.com/fatih/status/980918398417817600

[briansmith]

In case the Tweet gets deleted, it's just putting this into the main module and removing the CLI arguments:

func init() {
    flag.Set("logtostderr", "true")
}

[klingerf]

I'd like to propose that we start sending all glog message to /dev/null instead of writing them the stderr. I've only ever seen glog messages printed at controller startup before the proxy is initialized. These messages are harmless but can be alarming. I think it would better to suppress them altogether, and rely on our own logging implementation to tell us if there's a problem.

[klingerf]

Note that we can still default sending glog message to /dev/null using the flag.Set approach that's suggested in the description for this issue.

[klingerf]

We should enable glog logging to stderr if -log-level=debug, otherwise we should suppress glog logs.

[klingerf]

I took some time to read the glog source, and unfortunately it appears that it's not possible for us to disable glog logs on a readonly filesystem. This relates to kubernetes/kubernetes#61006. Glog writes its logs to either stderr or to a directory on disk, and we have to write to stderr since our processes are running inside scratch containers and can't write to disk.

I'll still use this issue to track overriding glog's default settings to log to stderr automatically, which will allow us to remove the -logtostderr flags from our install config.