[Prototype] Node 24: count await continuations that block on their dispatch fiber (port of #12)

[jbaczuk-qualia]

Summary

Stacked on #9 (the Node 24 port). Same change as the Node 18 version, #12. Adds Fiber.awaitContinuationBlocks, a process-wide count of code after an await that blocked on the hidden fiber the dispatcher gave it (Future.wait, Promise.await, a sync Meteor call). On stock fibers each of these throws "Can't wait without a fiber", so the count shows where code still depends on fibers after an await. Code that just runs on a dispatched fiber and never blocks behaves the same as on stock, so it isn't counted.

File	Change
fibers_async.js	Incremented in the dispatcher's existing "job blocked" branch (reuse dispatcher: run() !== kIdle and the fiber is still the dispatch fiber; one-off dispatch: the job didn't finish). Counted once per blocking job. Ordinary yields and jobs that don't block pay nothing. Stored on the native module, so every copy of fibers in a process shares one count.
test/await-continuation-blocks.js	Non-blocking awaits → 0; one continuation blocking twice → 1; two blocking continuations → 2; awaits outside a fiber → 0. Prints pass (skips) when dispatch is off.

The labelled metric and the dev-only call-site recorder live in qualialabs/qualia#57387.

Testing

Not run on Node 24: the code is identical to the Node 18 version (cherry-picked; the only conflict was next to the AsyncContextFrame probe). Results from the Node 18 build:

• test/await-continuation-blocks.js: pass by default, with FIBERS_AWAIT_REUSE=0 (one-off dispatch path), and with FIBERS_AWAIT_DISPATCH=0 (skips).
• Microbenchmark: 5,000 blocking continuations × 12 runs → Fiber.awaitContinuationBlocks = 60,000, an exact count.
• In local qualia the count matched prom-client's labelled metric and the dev-only recorder (4/4 and 3/3).

🤖 Generated with Claude Code