report extra gpu device when config one enabled_vgpu_types

That's because we don't verify the sections if we only have one reported type (and then we create inventories for all the GPUs)

For fixing it, you can use a fake vgpu type and create a new section for it, like :

enabled_mdev_types = nvidia-235,fake
[mdev_nvidia-235]
device_addresses = 0000:84:00.0,0000:85:00.0

[mdev_fake]
device_addresses = 0000:00:00.1,0000:00:00.1

That said, we could fix this issue by still verifying the section even for one reported type, but just not returning an exception if there is no section.

Related fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/nova/+/814743

This is also specifically affecting MIG-enabled nvidia GPU:

in MIG case, the card is represented as 16 VFs due to SR-IOV, but in fact can not support more that 7 of the smallest possible mdev type.
If I want to expose only one VGPU type from compute, nova will create and report all 16 VFs as resource providers to placement, leading to resource capacity over-report.
Also, when using single mdev class, one can not set a custom resource class name for it, only default 'VGPU' resource class.

That then requires an extra dance with resource provider traits that could've been avoided if we just properly parse those dynamic config sections if they exist for the case of single mdev type enabled.