Data Driven testing for BeamSQL

Current way to write BeamSQL test cases is too heavy: developers need to initialize pipeline, deal with PCollection, and use PAssert to verify pipeline results (sometime through INSERT INTO table and read data from table for assertion).

Data driven testing, instead, should only ask developer to provide SQL query and a expected result in the form of List (simulate rows from result table). The test execution interface should just be a static function like "List run(String query)", and returned rows can be compared with expected result by checking equality.

Imported from Jira [BEAM-7142](https://issues.apache.org/jira/browse/BEAM-7142). Original Jira may contain additional context.
Reported by: amaliujia.