Flavor/Image-based limit for virtio-net multiqueue

Nova currently supports enabling virtio-net multiqueue, where the number of queues is automatically set to match the number of guest vCPUs. While this behavior works well in many cases, it can be suboptimal for instances with a large number of vCPUs, particularly in OVS-DPDK based environments, where an excessive number of queues may increase CPU overhead and reduce dataplane efficiency.

Nova provides a `max_queues` configuration option in `nova.conf` to limit the number of queues. However, this option is configured per compute node and applies uniformly to all instances running on that node. As a result, operators cannot easily tune virtio-net multiqueue behavior based on individual instance sizing or workload characteristics.

This blueprint proposes adding support for limiting the maximum number of virtio-net queues using flavor extra specs or image metadata when multiqueue is enabled. The limit is fixed at instance creation time and allows more flexible, workload-aware tuning, while still respecting the upper bound configured via `nova.conf` on each compute node.