Global resources are never cleaned

We use a global DNS `ServiceDiscoverer` for clients when no `ServiceDiscoverer` is provided by the user. This DNS `ServiceDiscoverer` uses `DefaultDnsServiceDiscoverer` which requires closing, failing which leaks a channel.

EDIT: `GlobalExecutionContext` contains an `IoExecutor` and `Executor` that are also never closed.

Start by tracing the global DNS ServiceDiscoverer and DefaultDnsServiceDiscoverer, then inspect GlobalExecutionContext and its IoExecutor and Executor. Determine how these global resources are owned and where shutdown is expected; done means the DNS channel and execution resources are closed without disrupting clients that use the defaults.

Written by the indexing model from the issue text.