Use Promises for async api calls

[will]

The first try at manually using spawn and channels worked, but if there
was an exception raised, it'd print a stack trace like this instead of
being handled nicely

Unhandled exception in spawn:  (CB::Client::Error)
  from src/cb/client.cr:517:5 in 'exec'
  from src/cb/client.cr:484:5 in 'get'
  from src/cb/client.cr:144:5 in 'get_teams'
  from src/cb/completion.cr:126:22 in '->'
  from /usr/local/Cellar/crystal/1.4.1/src/fiber.cr:146:11 in 'run'
  from /usr/local/Cellar/crystal/1.4.1/src/fiber.cr:98:34 in '->'
Unhandled exception in spawn:  (CB::Client::Error)
  from src/cb/client.cr:517:5 in 'exec'
  from src/cb/client.cr:484:5 in 'get'
  from src/cb/client.cr:144:5 in 'get_teams'
  from src/cb/client.cr:234:18 in 'get_clusters'
  from src/cb/completion.cr:128:22 in '->'
  from /usr/local/Cellar/crystal/1.4.1/src/fiber.cr:146:11 in 'run'
  from /usr/local/Cellar/crystal/1.4.1/src/fiber.cr:98:34 in '->'

The way to get around this would be to have the channel also be able to
take an exception, and resuce exceptions in the spawn block and send
them on the channel, then raise them in the main thread. This is a lot
of boilerplate work. I started writing a thing to do this for us, then
realized it was basically the promise pattern, and that someone else
probably has done that. I was a little hesitant to pull in this library
but it's not that large, and seems to do a good job.

So now when there is an exception we get the nicer output since this
takes care of all the steps to reraise exceptions in the right spot:

error: 404 Not Found
       GET to /teams
       message: Resource 'account' with ID 'wef' couldn't be found.
       request_id: 26ac03b2-159e-4f0f-809e-84c474fe7b79

But then I realized that you can't actually call get clusters and get
teams at the same time, since you need to know the teams in order to
list their clusters. And this whole time for completion we were making
an extra call to get teams, since get_clusters with no args makes it's
own call to get teams. So I removed that unnecessary call from the
completion. The unhandeled exception problem though is a potential
problem in get_clusters, so it makes sense to use the new Promise
library there I think.

[abrightwell]

completion_spec needs something like

  def get_clusters(teams : Array(Team))
    [Cluster.new("abc", "def", "my cluster", [] of Cluster)]
  end

  def get_clusters(team_id : String)
    [Cluster.new("abc", "def", "my cluster", [] of Cluster)]
  end

For the tests to pass.