Ability to add a new sled to the lrtq cluster

[andrewjstone]

A sled that is plugged into an existing initialized rack needs to be told it is a learner node so that it can learn its key share. The bootstrap agent has access to a NodeHandle that can be used for this purpose.

Once the learner has learned its share it can ask for other peers' shares so that it can recompute the rack secret. This initialization happens in request_sled_agent, which is triggered via RSS or having an already ledger persisted SledAgentRequest on cold boot. However, new sleds will not have this, and it is likely we need to plumb a sled addition request all the way from the UI to perform this initialization.

[jmpesp]

Related to #2545?

[andrewjstone]

Related to #2545?

Indeed! LRTQ is very unsophisticated in that if a new sled learns a share, you can't really revert that without wiping its cluster directories on the M.2s, which I suppose could be the undo part of that saga action. You also have the relatively minor issue that the sled's baseboard will be permanently marked as using a share that was handed out by a sled. There's no way to revoke that. This is all temporary though, until a trust quorum protocol supporting reconfigurations and rekeying is implemented.

[andrewjstone]

This was completed in #4545