Implement NPD metrics mode

[xueweiz]

Here is the proposal.

This issue tracks the implementation of the proposal:

• Check-in the proposal to NPD project's README (Add metrics mode proposal to README #285)
• Refactor problem daemon registration and initialization, so that they are more modular, and rely on less per-daemon initialization code duplication. (node-problem-detector: report disk queue length in Prometheus format #275)
• Refactor events/condition reporting into k8s_exporter. (node-problem-detector: report disk queue length in Prometheus format #275)
• Add a SystemStatsMonitor problem daemon, to monitor some system metrics; and add an optional Prometheus metrics exporter. (Update dependencies in Godep for #275 #286 node-problem-detector: report disk queue length in Prometheus format #275)
• Implement feature in System Log Monitor and Custom Plugin Monitor, to allow them to report metrics. (Report metrics from system-log-monitor #300 Report metrics from custom-plugin-monitor #315 )
• Implement option for disabling problem daemons at compile time. (Allow compilation time disabling for each type of Problem Daemon. #294 )
• Implement option for disabling exporters at compile time. (Current exporters includes a k8s exporter and Prometheus exporter. Since they are commonly used, compile time disabling is not high priority for now. We can re-prioritize if the situation changes.)

[wangzhen127]

/cc @wangzhen127

[xueweiz]

/cc @Random-Liu @jiayingz @dchen1107

[xueweiz]

/assign @xueweiz

[andyxning]

/cc @andyxning

[xueweiz]

/cc @kewu1992

[xueweiz]

Now that #315 is merged, this issue is finally complete! It will be release in NPD v0.7.0

Huge thanks to Zhen, Lantao and Andy for the reviews and feedback!

There are a few follow up improvements, most importantly, e2e tests for NPD (#296). I really want e2e test to happen, so that we can have more confidence when changing NPD code.