[NBS] Add VFIO-PCI passthrough support for local NVMe devices in QEMU VMs

[sharpeye]

Currently, NBS can expose local NVMe disks to QEMU VMs via the vhost-user protocol. However, vhost-user has significant drawbacks: it introduces performance overhead that prevents full utilization of disk capabilities, and it doesn't support NVMe-specific features like NVIDIA GPUDirect Storage (GDS).
To address these limitations, NBS should support exposing local NVMe disks via VFIO-PCI passthrough. This would eliminate the virtualization overhead, enable maximum I/O performance, and provide full access to advanced NVMe features required by GPU-accelerated workloads.