In playbooks, hosts can be a YAML list. We templated the list before
converting it to a semicolon-separated string, which actually templated its
repr. This converts to a string first. A basic unit test is included.

Do not run unit tests against the EVENTS code because it is subject to change and that does not mean anything is broken.
If we want to test playbooks, it's best to test the result of a playbook to make sure it does the correct thing.  Also
remove playbook2/3 yml which were not used.

There's not need to run the complete include through the templating engine.
Several variables were not included before the conditional was evaluated.

The original patches should have conflicted?
53bde0bf vs efde61e5

Also change test name collision.

Add tests for new advanced inventory features (groups of groups, group variables) in the default INI format file.

Next up: YAML format.

format compatible with this still, and add some tests for INI-style groups of groups
and variables.

This is a list of all the groups a host is in.

ansible-playbook release-my-app.yml --extra-vars="version=123"

And make $version available in the playbook without re-editing the file

If the module result contains "ansible_facts", that will be added to the setup
cache.

Don't force down ansible facts back to setup, the setup module won't like parsing them on input and that
data is already there.

The fetch module really should preserve the whole directory structure being fetched to allow subsequent calls,
particularly in playbook, to recreate the host tree structure.  Making it thus.

Also account for path difference in sleep in different distributions.

See test/yaml_hosts for an example.

Hosts can be part of multiple groups.

Groups can also have variables, inherited by the hosts.
There is no variable scope, last variable seen wins.

Simplify playbook tests so things are not timing dependent in the poll section, which is leading to tests
not being consistent between runs, even though there wasn't an error.  Now we'll just check the final change
counts, which should be just as solid and lead to less churn in the events file.