Make integration test infrastructure smarter w.r.t. Cassandra versions

[absurdfarce]

Bug Report

Description

The PR for #106 makes the existing integration tests more granular by creating distinct tests for stdout and output-dir outputs for both CQL and NB exporters. Overall this is a goodness as it allows each test to be granular and focus on a single thing. But it has exposed a significant shortcoming w.r.t. how the integration test framework manages Cassandra versions.

Specifically, the current framework iterates over each Cassandra version under test for each individual test case. We therefore wind up doing C x T C* starts (each of which has some built-in wait time for the server to come up and be ready) for C C* versions and T test cases. This is okay right now (since we only have a small number of integration tests and the C* version pool is fairly constant) but doesn't scale at all.

The test framework should be modified to (a) start up each C* version only once and (b) loop through each test case once the server is ready. This would greatly minimize our delays based on C* startup. This does introduce some significant change to our current assumptions, however. Specifically our current test cases assume a pristine environment so they don't worry about dropping existing keyspaces that might be present before starting the test. Some experimentation was done around this when working on #106, however, and it isn't hard to manage at all.

┆Issue is synchronized with this Jira Task by Unito
┆Issue Number: AD-34

[absurdfarce]

Worth considering how this work should interact with #110