Merge rancher-operator into rancher

Embed channelserver in rancher

[master] Implement deterministic naming for Bindings to mitigate duplicate issues

[master] Update apiGroups to use explicit group

Do not set unknown condition for creating cluster

Without this change, if a GKE cluster creation fails for some GKE-side
reason, the state of the cluster object will appear to rapidly flip
between provisioning and failed, and the Rancher logs will be filled
with a constant stream of messages:

  2021/04/13 10:15:43 [ERROR] error syncing 'c-tcc2p': handler gke-operator-controller: failed setting cluster [] condition Provisioned false with message: creation failed for cluster cluster-name, requeuing

This is a confusing UI behavior and the volume of logs emitted has the
potential to generate large log files unexpectedly quickly.

This change removes the initial status update during the "creating"
phase. The Provisioning condition will be set a few lines later to
either Unknown or False, so the initial status update request is
superfluous. Removing it prevents a case where the cluster object might
be have been modified and the condition status cannot be set, which
causes the onClusterChange handler to error out without calling
`e.ClusterEnqueueAfter(cluster.Name, enqueueTime)` and so causes the the
handler to re-run repeatedly without a delay.

The problem was observed for GKE clusters but this fix applies to both
EKS and GKE.

This change also fixes the SetUnknown/SetFalse/SetTrue helper functions
to return the update error if there is one, instead of returning an
error string with the context of the error but not the error itself, so
that when update conflicts happen it's no longer hidden.

(cherry picked from commit b44c2232)

Add ACE support for Rancher provisioned RKE2/K3S clusters

Add Imported field to GKE upstream spec builder

[master] rework nodepools to use cache during update

Trim chart label to 63 characters in helm chart

Use mirrored images for workload test

Validation - Update test image repo

Without this patch, if a GKE cluster is imported to Rancher, and then a
change from GKE is synced back to the cluster in Rancher, "imported"
will flip back to false, and the guard in gke-operator that prevents
sending a DELETE request to imported clusters that are removed from
Rancher gets thwarted. This change fixes the upstream spec builder
function in the cluster refresher to always use the config spec value
for the Imported attribute.

(cherry picked from commit 9373c677)

Rely on Windows manifests instead of manual tags

A previous PR (31548) added support for Google cloud credential.
However, this made the unsupported Google node driver active. In order
to deactivate the node driver and create the cloud credential schema for
the Google driver, a new field is added to the NodeDriverSpec:
AddCloudCredentialByDefault. If this is `true` for any node driver, then
the driver is downloaded, if necessary, and the cloud credential schema
created. This would work for any node driver, not just the built-ins.

[master][forwardport] Whitelist eks and kms CN domains

[master] Fix CIS scan when PNI is enabled

vendor update lasso

[master] Take apiGroups into account when processing auth objects

Bump helm unittest to support KubeVersions