fix: check group map before enable pci device passthrough

[WebberHuang1118]

Problem:
PCIDevice's device plugin uses an out-of-date group after node rebooting

Solution:
check the group map before enabling PCIDevice device plugin

Related Issue:
harvester/harvester#6892

Test plan:

• Enabling a PCIDevice
• Creating a VM attach the PCIDevice
• After starting the VM successfully, rebooting the node
• Checking if the VM can start successfully across node rebooting

[ihcsim]

For my own understanding, if the device's status.iommuGroup is stale, should it be updated to match that found in the group path, like what the pci device controller is doing in

pcidevices/pkg/controller/pcidevice/pcidevice_controller.go

Line 98 in c65a093

devCopy.Status.Update(dev, nodename, iommuGroupMap) // update the in-memory CR with the current PCI info

Looks like the pci device controller is also doing the same group map check in

pcidevices/pkg/controller/pcidevice/pcidevice_controller.go

Lines 47 to 52 in c65a093

	// Build up the IOMMU group map
	iommuGroupPaths, err := iommu.GroupPaths()
	if err != nil {
	return err
	}
	iommuGroupMap := iommu.GroupMapForPCIDevices(iommuGroupPaths)

[ibrokethecloud]

@ihcsim the issue is when the iommu group is stale, likely after an OS upgrade, the device plugin gets setup with the wrong iommu group, which results in wrong iommu group being passed to the launcher pod.

With this change we wait for the periodic device reconcile to update the pcidevice object and this blocks an incorrect device plugin from being setup.

[ibrokethecloud]

lgtm. thanks.