Fix global partitioning with Separated handler mode

[jeremydmiller]

Summary

• Fixes NoHandlerForEndpointException when combining global partitioning with MultipleHandlerBehavior.Separated
• Extends the fanout handler logic in HandlerGraph.HandlerFor(Type, Endpoint) to also activate when the incoming local queue is part of a sharded/global partition topology (UsedInShardedTopology = true)
• Adds integration test with two global partitions, multiple handlers per message type, and cascading messages

Problem

When global partitioning routes messages to companion local queues and Separated mode moves all handlers to sticky handler-specific local queues:

1. Message arrives at local://partition-a2/ (global partition companion queue)
2. No sticky handler matches this endpoint
3. Fanout handler logic checks endpoint is not LocalQueue — but it IS a LocalQueue
4. Falls through to throw new NoHandlerForEndpointException

Fix

The fanout handler should also be created when the incoming LocalQueue is part of a sharded topology (not a sticky handler's own queue). The UsedInShardedTopology property on the endpoint distinguishes partition queues from arbitrary local queues, preserving the existing NoHandlerForEndpointException behavior for misrouted messages.

Test plan

• New integration test passes: two global partitions, four handlers (two per message type), cascading message flow
• All 1161 CoreTests pass (1160 existing + 1 new), including get_an_explanatory_message which validates that NoHandlerForEndpointException is still thrown for arbitrary unhandled local queues

Supersedes #2292

🤖 Generated with Claude Code