Add --namespace flag support to workloads (#1196)

Allows specifying a target Kubernetes namespace for workload commands (create, list, delete, wait).
This avoids hardcoding the 'default' namespace and enables running workloads in custom namespaces.

Also updates setup_k8s_service_accounts to create service accounts and role bindings in the specified namespace.

Co-authored-by: Ankit Kinra <ankitkinra@users.noreply.github.com>

Check if Kubernetes resources exist before creating them in XPK setup (…

…#1195)

* Check if Kubernetes resources exist before creating them in XPK setup

In GKE clusters with restricted permissions, users might have read access to Roles/ServiceAccounts/RoleBindings but not create access. XPK was previously using try-create-catch-409 logic, which causes 403 Forbidden errors if the user lacks create permission, even if the resource already exists.

This change implements a 'check before create' pattern for ServiceAccount, Role, and RoleBinding creation in cluster.py. It checks if the resource exists first (using GET), and only attempts to create it (POST) if it doesn't exist. This allows administrators to pre-create these resources and lets developers run XPK without needing elevated creation permissions.

* Format cluster.py with pyink

Switch use-parallel-containers to be false by default (#1193)

* Switch use-parallel-containers to be false by default

* update goldens

fix: Fix --adapt-from-ct for externally managed jobset and kueue (#1192)

* feat: skip Kueue and Jobset manifest installs if managed by Helm

* Add unit tests for helm-managed Kueue and Jobset

* Refactor is_managed_by_helm to use jsonpath for speed and target deployments explicitly

* Simplify is_managed_by_helm to only check managed-by label

* Simplify code

* Rename out to container_image in _should_install_jobset

* Rename is_managed_by_helm to is_managed_externally

* Fix pylint trailing whitespaces

* Fix pyink formatting

fix: Fix gcluster and make Nightly Tests passing again (#1191)

* Remove secrets.CLUSTER_ARGUMENTS and always use CLUSTER_NETWORK_ARGUMENTS

* Fix ray cluster e2e

* Use the most recent gcluster version

* Update gcluster version in blueprint generator and docker manager

* Remove the "v" prefix from jobset version in blueprint_generator

* fix(blueprint): conditionally omit host_maintenance_interval for spot A3 Mega clusters

Preemptible VMs (spot instances) do not support periodic maintenance intervals. This caused cluster toolkit deployments using the A3 Mega blueprint with `--spot` to fail with a Google Compute Engine 'Invalid value' error. This commit conditionally omits `host_maintenance_interval` if the capacity type is SPOT.

* Fix GPU test and update test fixture to fix failing unit tests

Also format python code with pyink.

TAG=agy
CONV=a5a75b42-5457-49d9-bdbb-e84c6347b9b7

Update kubernetes dependency to version 35.0.0 (#1178)

Update the client so it supports HTTPS_PROXY

feat: support --device-type for xpk cluster create-ray (#1175)

Clarify path installation is only for installing from source (#1170)

fix: JobSet Status Resolution (#1161)

* Fix JobSet Status Resolution Regression

* Remove overly protective shlex escaping

Enable Crane (#1120)

* Enable Crane. Keep a golden and integration test for Docker (to be cleaned up later)

* Remove non-crane tests