Pass --single option to cloud runners

[0x2b3bfa0]

This pull request passes the newly introduced --single option to the launched cloud runners.

Closes #94, required by iterative/cml#476

[DavidGOrtega]

@0x2b3bfa0 is our wait mechanism aware of the error is going to throw if the runner is already in use?

[DavidGOrtega]

Another issue is that we need to deploy the machine before knowing this. We cant do anything else now but we should add a ticket as a technical debt.
Ne able to check single without launching the machine.

[0x2b3bfa0]

@DavidGOrtega, we're running the uniqueness checks on the cml-runner side. Should we move them so they run on the instance?

[DavidGOrtega]

@0x2b3bfa0 thats what I say. Right now would be a big effort change so would be better to make a ticket for that change just only to avoid spin a machine.

My concern right now is if the runner fails we are logging this correctly in the provider like the await does.

[0x2b3bfa0]

@DavidGOrtega, I suspect that we've unadvertently mixed up --single and --reuse in the previous comments after reading #94. The former argument is being passed to the cml-runner process on the cloud instance (as per this pull request), while the latter is being handled locally on the original non-cloud invocation.

If the runner fails and emits the JSON line we're expecting, the behavior should be similar to any other failure case. 🤔 Is that what you mean?

[DavidGOrtega]

lgtm