More Vector Set test hardening

[kevin-montrose]

Looking into some recent failures, and getting some (rare) local reproductions, these failures seem linked to WaitForReplicaAofSync.

My current theory is that the asynchronous replay of VADD and the background cleanup after DEL|UNLINK confuses this test function. Not all tests trigger it, but MigrateVectorSetWhileModifyingAsync seem pretty prone, so removing these calls which don't appear* strictly necessary.

• Waiting on a CI run (result: 2 failures)

Some null refs still happen after this change... looking at possible use-after-Dispose cases, also in replication.

Found a fixed a bug in AofProcessor.Dispose not shutting down VectorManager replication tasks.

• Waiting for CI run (result: pass)
• Stressing locally (result: occasionally seeing WaitForMigrationCleanup timeout, but Vector Set's don't appear to be at fault)

Doing a bit more digging into WaitForMigrationCleanup failures...

_{*These tests grew from existing cluster tests, so the "wait" calls were not added with any rigor.}

[Copilot]

Pull request overview

This PR aims to reduce flakiness in Vector Set cluster tests by removing WaitForReplicaAofSync calls that appear to be unreliable when VADD replay and post-DEL|UNLINK cleanup are asynchronous.

Changes:

• Removed WaitForReplicaAofSync waits from MigrateVectorSetWhileModifyingAsync after slot migration and before post-write validation.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.