Autoscaling from 0 workers

[net-expertise]

Hi,
When no worker bears the tag defined in a task, pipelines don't pass run checks and the following error is returned:
run check: no workers satisfying: resource type 'registry-image', tag 'dummy', version: '2.4'

It would be nice to have the ability to send tasks in the waiting tasks for later execution.
The idea behind it is to autoscale from 0 to N workers

As of now, combining the "limit-active-task" strategy and concourse_steps_waiting as a custom metric works great in a Kubernetes environment. We can (up/down)scale just fine. Tasks go in the waiting queue which triggers the creation of an additional worker, if the 1st worker already has a running task (with a task limit of 1 for ex). But 1 worker always needs to be up and running, even though no task is running.

There could be an environment variable such as CONCOURSE_NO_COMPATIBLE_WORKER_ACTION = Fail/Queue

Additional thought: I have no idea if there's a timeout on waiting tasks?