Move dead-silo message break to connection maintainer

[ReubenBond]

Part 5 of 7 split from #10085.

Problem:
Breaking outstanding messages to a dead silo was performed from LocalGrainDirectory/Catalog-related cleanup, even though it is unrelated to grain-directory state.

Solution:
Move BreakOutstandingMessagesToSilo handling to SiloConnectionMaintainer when a remote silo is marked Dead, keeping LocalGrainDirectory focused on directory reconciliation.

Stack:
Merge after #10089. This branch is stacked on split/pr10085-04-processing-cleanup; until earlier PRs merge, GitHub may show earlier stack changes. Incremental compare: ReubenBond/orleans@split/pr10085-04-processing-cleanup...split/pr10085-05-dead-silo-message-break

Review focus:
Dead-silo message break placement and LocalGrainDirectory/Catalog separation of concerns.

[Copilot]

Pull request overview

This PR continues the directory/membership cleanup split by relocating “break outstanding messages to dead silo” into the networking layer (SiloConnectionMaintainer) and further refactoring LocalGrainDirectory to reconcile membership from IClusterMembershipService snapshots instead of status-oracle event plumbing.

Changes:

• Move BreakOutstandingMessagesToSilo invocation to SiloConnectionMaintainer when a remote silo is marked Dead.
• Refactor LocalGrainDirectory to process membership snapshots/updates, adjust local directory/cache via IsDefunctActivation, and take over terminating-silo activation deactivation logic previously in Catalog.
• Update handoff/retry behavior and add targeted unit tests for defunct-activation determination.

Show a summary per file

File	Description
test/Orleans.Runtime.Internal.Tests/LocalGrainDirectoryTests.cs	Adds unit coverage for LocalGrainDirectory.IsDefunctActivation behavior across silo statuses/unknowns/successors.
test/Orleans.Core.Tests/Directory/CachedGrainLocatorTests.cs	Updates test wiring for LocalGrainDirectory’s new IClusterMembershipService dependency.
src/Orleans.Runtime/Networking/SiloConnectionMaintainer.cs	Breaks outstanding messages and closes connections when a remote silo is marked Dead.
src/Orleans.Runtime/GrainDirectory/LocalGrainDirectory.cs	Moves to snapshot-driven membership processing; centralizes terminating-silo activation deactivation and defunct-entry cleanup.
src/Orleans.Runtime/GrainDirectory/ILocalGrainDirectory.cs	Removes the IsSiloInCluster method from the internal interface surface.
src/Orleans.Runtime/GrainDirectory/GrainDirectoryHandoffManager.cs	Adjusts successor checks, filters terminating addresses during handoff acceptance, and changes pending-operation retry/dequeue behavior.
src/Orleans.Runtime/Catalog/Catalog.cs	Removes catalog’s silo-status-change directory cleanup/break-outstanding-message responsibilities.

Copilot's findings

Comments suppressed due to low confidence (1)

src/Orleans.Runtime/GrainDirectory/LocalGrainDirectory.cs:677

• Same as RegisterAsync: hopCount > 1 skips the retry delay/re-check on the first forwarded hop (hopCount==1), despite the comment saying this should happen after the first forward and other operations using hopCount > 0. This can lead to rapid multi-hop forwarding when ownership is in flux. Consider using hopCount > 0 here (or align all forwarding methods to the same semantics).

            // After the first forward, we insert a retry delay and recheck owner before forwarding again
            if (hopCount > 1 && forwardAddress != null)
            {
                await Task.Delay(RETRY_DELAY);
                forwardAddress = this.CheckIfShouldForward(address.GrainId, hopCount, "UnregisterAsync");

• Files reviewed: 7/7 changed files
• Comments generated: 2

[ReubenBond]

Closing as obsolete: the later stacked PR #10092 has already merged and its merge commit includes the #10090/#10091 changes, including the xUnit logger crash fix. Current main also includes the follow-up directory retry PRs #10094 and #10095.