Right now this only provisions the genesis node. We need to implement the retrieval of the genesis
IP and peer ID before we can run against the remaining nodes.

It uses a very similar pattern to Terraform, with an interface, so that we can do behaviour-based
unit testing.

There were over 700 lines of tests and setup code inside a single `test` module, which was becoming
unwieldy to work were.

This splits the setup code into one module, then a module for each function on `TestnetDeploy` that
is under test. With the use of modules, the function name was removed from the test case names.

The smaller functions are easier to test and don't require as much mocking setup.

The multiaddr of the genesis node is composed by running `ansible-inventory` to get its IP address,
by running the rpc client to get its peer ID.

After we have that, we can then run against the remaining nodes.

Those two functions were separated for easier test setup and admin.

It's possible that before we run Ansible, the SSH service is not yet running on the genesis node.
This is something we seen happen occasionally, so we added in a little loop to sit and wait for it
to become available.

We only check this for the genesis run because the remaining nodes are provisioned after genesis, by
which point the SSH service should be available.

This will control how many node processes run on each VM. The default is 20.

If the `--branch` and `--repo-owner` arguments are used, the build machine will be spun up to build
a custom binary, which will then be uploaded to S3. Then when the nodes are being provisioned, this
binary will be pulled in.

This creates the `.env` file by gathering input from the user, trying to assist them with which
values should be provided.

We are leaving it to the user to setup Terraform and Ansible on their OS.

This still needs to be extended with some information on how to obtain the IP addresses of each
VM.

This will produce a list of all the VMs participating in the testnet, along with the SSH user for
the cloud provider.

Addresses two issues that were noted during review:

* Provide more helpful error message when inventory list is empty
* Set other environment variables for Digital Ocean based on the `DO_PAT` variable

For the first issue, there are a few different ways in which it might be possible for the
environment to be non-existent. These were documented with comments in the code, which you can see
in the diff.

In addition, license headers were applied where they were missing.