…e" error

E2E tests are flaky - occasionally failing during cluster setup with the following error:
```
❌ Kai Scheduler installation failed on attempt 1/3: helm install failed:
    the server could not find the requested resource

❌ Kai Scheduler installation failed on attempt 2/3: helm install failed:
    cannot re-use a name that is still in use

❌ Kai Scheduler installation failed on attempt 3/3: helm install failed:
    cannot re-use a name that is still in use
```

* Root Cause
When a Helm install fails partway through (e.g., due to a race condition where the Kubernetes API server isn't fully ready), Helm leaves a release record in a failed/pending state. The existing retry logic attempts helm install again with the same release name, but Helm rejects this with "cannot re-use a name that is still in use" - even though the release is in a failed state.

* Solution
Add Replace = true to the Helm install client configuration. This allows Helm to replace releases that are in a failed/pending state, enabling retries to work correctly.

* Testing
Ran e2e tests multiple times to verify the fix handles the race condition
The Replace flag only affects releases in failed/pending states, not successful releases