fix: drain accepted request work before service shutdown

[pjb157]

Summary

• keep analytics and response writers alive until Axum finishes in-flight requests
• track and join work accepted by background=true requests before persistence shutdown
• flush Outlet request/response capture and handler queues before cancelling writers
• pin the reviewed Outlet lifecycle commits without using workspace patches

Incident mechanism

On SIGTERM, the control layer cancelled shared background services before accepted request and post-response persistence work had completed. That closed analytics/response-writer channels while requests were still draining, producing the observed channel closed failures.

Dependencies

Merge in order:

1. feat: add graceful request logger drain outlet#99
2. chore: use outlet graceful drain support outlet-postgres#83
3. this PR

The git revisions make this branch reproducible before crate releases; they can be replaced with released versions in normal dependency maintenance.

Verification

• CARGO_PROFILE_DEV_DEBUG=0 CARGO_INCREMENTAL=0 just lint rust
• DATABASE_URL=postgresql://localhost/dwctl_graceful_drain_20260731 RUST_TEST_THREADS=4 CARGO_PROFILE_TEST_DEBUG=0 CARGO_INCREMENTAL=0 just test rust
  • dwctl: 1917 passed, 2 ignored
  • all workspace unit, integration, and doc tests passed
• targeted graceful shutdown ordering test passed
• full Outlet and outlet-postgres test/lint suites passed in their prerequisite PRs

[cloudflare-workers-and-pages]

Deploying control-layer with    Cloudflare Pages

Latest commit:	eed501b
Status:	✅  Deploy successful!
Preview URL:	https://530b0a9f.control-layer.pages.dev
Branch Preview URL:	https://peter-graceful-http-drain.control-layer.pages.dev

View logs

[Copilot]

Pull request overview

This PR fixes graceful shutdown ordering so persistence-related background services (analytics, response writers, Outlet handlers) remain available while Axum drains in-flight requests, and it ensures background=true “accepted” work is joined before persistence shuts down. It also pins outlet / outlet-postgres to specific git revisions to pick up the lifecycle/shutdown behavior needed for the fix.

Changes:

• Introduce an Outlet shutdown handle and a cancel_after_drain helper to flush/join accepted work and then cancel background services only after HTTP draining completes.
• Track detached “accepted” work via a tokio_util::task::TaskTracker, and route background=true tasks through that tracker.
• Pin outlet / outlet-postgres dependencies to specific git revs and update the workspace lockfile accordingly.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 3 comments.

File	Description
dwctl/src/lib.rs	Refactors router construction to return an optional Outlet shutdown handle; adds accepted-task tracking and enforces post-drain shutdown ordering.
dwctl/src/inference/middleware.rs	Routes detached background request work through a shared TaskTracker so it can be drained on shutdown.
dwctl/Cargo.toml	Enables tokio-util runtime feature and pins outlet / outlet-postgres to git revisions.
Cargo.lock	Records the git-sourced outlet / outlet-postgres dependencies (and transitive updates) for reproducible builds.