Stop Redundantly Serializing Index Name in ReplicationRequest

[original-brownbear]

If we have the shard id we don't need to separately write the index
name. Especially for large bulk requests this wastes a non-trivial
amount of memory.

[elasticmachine]

Pinging @elastic/es-distributed (:Distributed/CRUD)

[DaveCTurner]

I was puzzled why ReplicationRequest has to deal with the case that it doesn't have a shard ID. I think that it's just that we use DeleteRequest and IndexRequest on both sides of the coordinating node, and the coordinating node fills the shard ID in itself. If so, can we instead separate the client-level usage from the shard-level usage and avoid this split personality?

[original-brownbear]

If so, can we instead separate the client-level usage from the shard-level usage and avoid this split personality?

Looks like you're right :) This actually seems like the better option. I'll try to see if I can find a way to make this happen without the BwC concerns around this becoming too crazy.

[original-brownbear]

Closing in favor of the more general improvements to this in #56209