watcher: log all watcher errors, make registry-fetch loop more resilient

[syphar]

we currently have an issue where the web server load leads to the db connection pool being exhausted, and some server errors being returned:

( I will fix that separately).

But: this lead to our watcher stopping to fetch new crates, without any error I saw.

This was a combination of two things:

1. we just spawned the watch_registry function as tokio task, but we never logged any errors.
2. errors in queue.is_locked or Index::from_config lead to just returning from that method, and stopping to check the registry.

This starts to log the errors from 1), and makes our registry loop more resilient so we would retry in these cases.

[syphar]

r? @GuillaumeGomez

[GuillaumeGomez]

Thanks!