bug: cache probe still performed even if count == 0

A common practice I've found for the `envbuilder_cached_image` resource is the following:

```
resource "envbuilder_cached_image" "cached" {
count = data.coder_workspace.me.start_count
...
}
```

However, when `start_count` is `0` I've still observed the resource being re-created.
This results in workspace shutdowns taking longer than they need, especially for large images.

Locate the envbuilder_cached_image resource and trace how a count of 0 is handled during planning or recreation. Reproduce the issue with start_count set to 0, then verify that no cache probe or resource recreation occurs and add or update coverage for that case.

Written by the indexing model from the issue text.