fix(atlas-service): use metricsId for clusterId when doing automation agent requests

[gribnoysup]

One thing I didn't notice when testing this: only serverless should be passing the uniqueId to the backend, other clusters use the metricsId. Also some drive-by fixes:

• sandbox electron proxy logs were not piped anywhere
• polling for connections in web was not working right because we never cleaned up the in-flight promise
• default backend config was not applying correctly in sandbox because the env value was missing