Some tiny fixes and improvements for kube backend + dask clusters.

[JSKenyon]

This PR includes:

• Propagation of environment variables specified in kube backend options into the dask_cluster worker and scheduler pods. This is needed to support e.g. NUMBA_CACHE_DIR when using a dask_cluster.
• Fix for a log message which was causing crashes as it referenced a potentially undefined variable.
• Replaces the dockerhub busybox images with the quay.io equivalents to circumvent rate limits (which come into play when pulling images onto hundreds of nodes).

[o-smirnov]

Great, thanks!