Fix: NFS directory creation and add NFS server reachability validation

[SAYUK09]

Summary

This PR fixes a critical bug where NFS directories (ctld_track, openldap, slurm, slurm_backups) were not being created during Slurm/Kubernetes provisioning, and adds proactive validation to check NFS server reachability before provisioning begins.

Changes

1. Fix NFS directory creation bug - The mount_config role was not actually mounting NFS shares on the OIM, preventing directory creation. Added mount_config_mode: "oim-mount" to:

• provision_slurm.yml - mounts NFS for Slurm config directory creation
• provision_kubernetes.yml - mounts NFS for K8s config directory creation
• mount_config/tasks/main.yml - skip cloud-init generation when in oim-mount mode

2. Fix OpenCHAMI URL variable collision - The validate_provisioning role was using cluster_name which gets overwritten by slurm_config (from openldap to slurm_cluster), causing DNS resolution failures.
   Changed to use stablehostvars['localhost']['oim_node_name']and domain_name instead.

3. Fix incorrect variable name - Changed orchestrator_input_dir to omnia_project_input_dir in generate_inventories/tasks/main.yml for HA config file path resolution.

4. Add NFS server reachability validation - New validation task that:

• Reads storage_config.yml and omnia_config.yml
• Resolves nfs_storage_name references from enabled features (slurm/k8s)
• Extracts NFS server IP from mount source field (e.g., 10.49.2.200:/mnt/test-nfs)
• Pings the NFS server to verify reachability
• Fails early if NFS server is unreachable, preventing provisioning failures
• Skips local devices (/dev/, UUID=, LABEL=) and CIFS shares (//)

5. Add missing feature flags to orchestrator state - Added slurm_support, ucx_support, and openmpi_support to prepare_orchestrator.yml so downstream roles can correctly check these flags.

Testing

• Verified NFS mount at /share_omnia is created and mounted.
• Verified directories are created: ctld_track, openldap, slurm, slurm_backups.
• Verified playbook runs to completion with zero failures.
• Verified sinfo on slurm controller node.