Draft: gpu-operator: Handle operator upgrades

• Problem: Currently operator reconcilation logic handles just creation of resources, but not updates. Thus, if operator is upgraded with existing resources, it would not apply changes as per ClusterPolicy spec. Also, all resources are deployed in different namespace than where operator is deployed. This can be avoided.
• Implementation: Handle update of all resources managed by operator. This allows upgrade of operator through OLM or Helm. Also, deploy all resources in the same namespace as operator for easy visibility and maintanance.
• Testing: Tested with operator deployment and upgrades and ensure reconcilation works.

Signed-off-by: Shiva Krishna, Merla smerla@nvidia.com