4.x: consider removing `x-cancel-on-ha-failover` consumer argument now that classic queue mirroring is gone

[michaelklishin]

As a follow-up to #9815, we can consider dropping x-cancel-on-ha-failover, a consumer argument that's rarely used in practice.

It is not entirely obvious to me, what the rabbit_channel of x-cancel-on-ha-failover should do now: always recover? For classic (no longer replicated) queues the choice is less obvious.