refactor(sinktools): use pin_project in LazySinkSource

[MingweiSamuel]

Stack: #2556

Kiro-generated, requires close review

[Copilot]

Pull request overview

This PR refactors sinktools::LazySinkSource to use pin_project with an internal pinned state machine and replaces the previous multi-waker approach with a dual AtomicWaker setup, while updating/expanding async unit tests around initialization driving.

Changes:

• Refactor LazySinkSource to a pinned SharedState enum using pin_project_lite (removing the old Rc<RefCell<...>> split-halves implementation).
• Replace MultiWaker (mutex + vec) with DualWaker using two AtomicWakers (sink vs stream) and a shared wake fanout.
• Update tests and add Tokio time dev-dependency feature to support timing-based assertions.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
sinktools/src/lazy_sink_source.rs	Reworks internal state/pinning and waker handling; updates and adds tests validating sink/stream-driven initialization.
sinktools/Cargo.toml	Enables Tokio time feature for dev-dependencies to support updated tests.

Comments suppressed due to low confidence (1)

sinktools/src/lazy_sink_source.rs:229

• poll_next returns Poll::Ready(None) when the initialization future resolves to Err, but it leaves the state in Thunkulating. If the stream is polled again after yielding None (which Stream contract allows), this will re-poll a completed future and may panic or otherwise violate the expectation that the stream remains terminated. Consider transitioning to a terminal state on init failure (e.g., an InitFailed/Terminated variant) so subsequent poll_next calls reliably return Ready(None) without polling the init future again.

                Poll::Ready(Err(_)) => {
                    return Poll::Ready(None);
                }

---

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

[Copilot]

This test now relies on a fixed wall-clock sleep (sleep(Duration::from_millis(10))) to wait for task wakeups/propagation, which can be flaky under slow or loaded CI runners (and unnecessarily slows fast runs). Prefer an event-driven wait (e.g., loop with yield_now() until the condition is true with an overall timeout, or a oneshot/Notify that signals when initialization completes) so the test is deterministic.

[cloudflare-workers-and-pages]

Deploying hydro with    Cloudflare Pages

Latest commit:	74e307b
Status:	✅  Deploy successful!
Preview URL:	https://2499bde7.hydroflow.pages.dev
Branch Preview URL:	https://lazy-2.hydroflow.pages.dev

View logs