gazctl: shards reassign subcommand

[jgraettinger]

If a shard fails during processing, it's status is set to FAILED and the relevant error is available through gazctl (bonus: make this more accessible; currently one has to use -o json / proto), but the shard isn't re-assigned away from the primary. This is deliberate, to "give" (force) the operator a chance to inspect and correct the fault.

Having done that, there isn't an easy way to re-assign / restart these shards short of deleting their current assigned pod.

This command should take an optional selector (if not provided, all shards are selected) and remove primary assignments for any shards which are currently failed. It'll probably also require a new consumer RPC to actually drop the Etcd assignment.