Add kubecluster cluster discovery

[jacobtomlinson]

Adds a discovery method for KubeCluster objects and also the from_name method required to reconstruct a KubeCluster object.

This PR works by serializing the KubeCluster object into a Kubernetes Secret resource, allowing it to be recreated at a later time.

This may not be an ideal approach, but opening this anyway as a reference.

from dask_kubernetes import KubeCluster

pod_spec = {...}  # Minimal pod spec omitted

cluster = KubeCluster(pod_spec, shutdown_on_close=True)  # Don't destroy the cluster if we destroy the object
cluster.scale(5)

del cluster

$ daskctl cluster list
NAME                        ADDRESS                TYPE         WORKERS  THREADS  MEMORY   CREATED   STATUS
dask-jtomlinson-bffca7fc-7  tcp://localhost:65125  KubeCluster        5        5  5.20 GB  Just now  Running

from dask_ctl import get_cluster

cluster = get_cluster("dask-jtomlinson-bffca7fc-7")

# Use cluster for Dask work

cluster.close()

Note the address is localhost as I used port forwarding in my config. This port will randomly change every time we run get_cluster (which is used within daskctl). For clusters using a LoadBalancer or NodePort this address will be consistent.

Depends on dask/distributed#4605

[jacobtomlinson]

Hopefully this can be replaced with something based on dask/distributed#5033 which would enable use of the scheduler as the key/value store instead of a secret.

[jacobtomlinson]

Closing in favour of #468