Discard fails on transient DNS errors

```
2024-02-05 18:38:35 Error discarding google:ubuntu-22.04-64 (feb051803-772047): cannot deallocate Google server google:ubuntu-22.04-64 (feb051803-772047): cannot perform Google request: Delete "https://www.googleapis.com/compute/v1/projects/snapd-spread/zones/us-east1-b/instances/feb051803-772047": proxyconnect tcp: dial tcp: lookup squid.internal: Temporary failure in name resolution
```

This should probably retry before failing and should also ensure the rest of the machines are properly deallocated if possible.

[charmcraft-spread.log](https://github.com/snapcore/spread/files/14170713/charmcraft-spread.log)

Start with charmcraft-spread.log and trace the discard/deallocation path for Google Compute Engine machines, focusing on the request that fails during transient DNS resolution. Reproduce or test the failure, then verify that retries occur and that remaining machines are still deallocated when one discard fails.

Written by the indexing model from the issue text.