Windows integration test don't raise errors on failure

### Description

Windows integration tests don't raise errors on failure

### Repro steps

Run current Github Actions:

```
Failed to download `melpa' archive.
Package refresh done
Failed to download `gnu' archive.
Setting `package-selected-packages' temporarily since "emacs -q" would overwrite customizations
Setting `package-selected-packages' temporarily since "emacs -q" would overwrite customizations
Package `s-' is unavailable
Importing package-keyring.gpg...
Importing package-keyring.gpg...done
Contacting host: melpa.org:443
Failed to download `melpa' archive.
Package refresh done
Failed to download `gnu' archive.
Setting `package-selected-packages' temporarily since "emacs -q" would overwrite customizations
Setting `package-selected-packages' temporarily since "emacs -q" would overwrite customizations
Package `s-' is unavailable
cask exec: error: Failed to execute buttercup -L . -L test --traceback full: [Errno 2] No such file or directory
Did you run cask install?
```

### Expected behavior

Action fail (red)

### Actual behavior

Action succeed (green)

Inspect the Windows GitHub Actions job and the `cask exec buttercup -L . -L test --traceback full` command shown in the report. Reproduce the integration test failure and trace whether its exit status reaches the action; done means the action is marked failed (red) when the test command fails.

Written by the indexing model from the issue text.