Build nightly end-to-end scenario test set to better find regressions

[clockfly]

We need a system way to assure quality for each release tag.

[clockfly]

Requirements:

End to end test, which means we should cover all use cases a normal user may encounter.

Fully Automated

• Commands under bin/ can run correctly. (in CoreSpec)(https://hub.docker.com/r/kairen/yarn-spark/))
• Read/write data from/to Kafka. (in ConnectorSpec)
• Submit/kill/query all built-in examples from command line shell. (in ExampleSpec)
• Message loss, machine crash, high availability test. (in StabilitySpec)
• Able to run Storm applications over Gearpump. (in StormCompatibilitySpec)

Manually or Out-of-scope

• Support Gearpump over YARN cluster. (in YarnIntegrationSpec use [Dockerfile]
• Apply configuration on-the-fly and check if it takes effects
• Submit/kill examples all built-in example from UI.
• Performance regression test.

[stanleyxu2005]

Here is a proposal