Failure during `influx setup` can leave system in an inconsistent state

`influx setup` isn't idempotent: if it fails after the CLI sends an onboarding request to the server, subsequent runs will fail with an "instance has already been set up" error.

The only thing the CLI does after the onboarding request is write a config file. We could potentially log/print the config contents on error (though it includes an auth token, so that could be a security risk).

Start by tracing the `influx setup` onboarding request and the subsequent config-file write, focusing on the error path between them. Determine a safe recovery or retry behavior that avoids exposing the auth token; done means a failed setup can be recovered from without an "instance has already been set up" error.

Written by the indexing model from the issue text.