warn when shotover doesnt know how to route a request

[rukai]

1

When shotover encounters a request type it does not know how to route it will route it to a random node.
This is a reasonable guess and in the worst case the broker will respond with a routing error which we return to the client.

However this does make it hard to track down issues caused by this random routing which can occur when

• an existing message type hasnt had routing implemented yet
• new message types are added to kafka

To make it easier to diagnose these issues this PR does two things:

• Adds explicit routing handling for all message types we currently cover in our integration tests
• Adds a warning if a request is unhandled and therefore routed to a random node.

2

Unfortunately things are never that simple.
The OffsetForLeaderEpoch request is being sent to shotover as part of our test suite and requires routing to partition leader but has no routing implemented yet.
It was failing due to incorrect routing but then eventually succeeds when the random routing arrives at the partition leader node.
So I decided to implement routing for this node rather than just adding explicit random node routing and leaving a TODO.

OffsetForLeaderEpoch requires request splitting for the same reasons that fetch and produce requests need to be split.
So this PR also implements routing, splitting and joining of OffsetForLeaderEpoch requests.

[codspeed-hq]

CodSpeed Performance Report

Merging #1779 will not alter performance

_{Comparing rukai:warn_on_unrouted_request (bb266ea) with main (26550b8)}

Summary

✅ 38 untouched benchmarks