[Jansen-w]

Summary

Retain task handles for removed durable disk-buffered sinks that continue draining after configuration reload, and include those tasks in the existing graceful-shutdown wait and deadline handling.

• Reload remains non-blocking and never waits for active retired sinks.
• Completed handles are synchronously reaped on later reloads without cancellation points; failures remain sticky so final shutdown reports false.
• Active retired disk sinks remain tracked until final shutdown or its configured deadline.
• Removed memory-buffered sinks keep their existing best-effort reload behavior; this change is intentionally limited to durable buffers.

References

Related: #26050

Vector configuration

Covered with topology regressions using an active sink and a removed sink with a blocking disk buffer.

How did you test this PR?

• Deterministic reload/shutdown drain regression, repeated 10 times
• Graceful-shutdown deadline expiry regression
• Reload-does-not-wait regression
• Completed-handle cleanup and failure-preservation regression, repeated 10 times
• Serial topology reload suite: 13 passed
• Transient-state and existing ordinary sink-removal tests
• cargo clippy --lib --tests -- -D warnings
• cargo fmt --all -- --check
• cargo vdev check changelog-fragments

The parallel reload suite retains an unrelated existing enrichment-state timing flake; the test passes serially and in isolation.

Is this a breaking change?

• Yes
• No

Does this PR include user facing changes?

• Yes. A changelog fragment is included.
• No. A maintainer will apply the no-changelog label to this PR.

Contributor Guidelines

• Please read our Vector contributor resources.
• After review begins, avoid force pushes so commit history remains reviewable.

[bruceg]

A couple of optional nits below; I wouldn't block on them.

[Jansen-w]

Closing this. As discussed with Rob, this likely needs an RFC to discuss what the expected behavior would look like. Some users may not expect/want the sink running continuously in the background.