mt76: use threaded NAPI

With threaded NAPI, the rx handler function is no longer bound to the CPU that
fired the interrupt, which significantly helps to spread the workload over
multiple CPUs, especially when multiple devices are using threaded NAPI at the
same time.
Exclude the tx handler from threaded NAPI by using a separate dummy netdev.
The work is small and short-lived enough that it makes more sense to run it in
softirq instead of creating a dedicated thread

Signed-off-by: Felix Fietkau <nbd@nbd.name>