Add `Runtime._connect()` method

Just like we have `Storage._connect()` to open a connection to the storage service, it would be useful to have a `Runtime._connect()` method too so the setup doesn't have to be repeated in every `.submit()` call. For example, with Kubernetes we have this at the moment:

```python
def submit(self, job_path):
kubernetes.config.load_kube_config(context=self.config.context)
client = kubernetes.client.ApiClient()
return kubernetes.utils.create_from_yaml(client, job_path)
```

The `client` could be kept in the object so it would only be created the first time a job is submitted. Maybe this doesn't need to be handled in the base class like for Storage, that's an implementation decision. Having a consistent way to do this between Storage and Runtime would be nice though.

Start by comparing the existing Storage._connect() pattern with the Runtime.submit() entry point and the Kubernetes example in the issue. Done means Runtime provides a consistent connection path and repeated submissions reuse the client instead of repeating setup; the implementation location and tests still need investigation.

Written by the indexing model from the issue text.