CloseableCursor implementation is racey

[Diggsey]

This is only used by the test suite, but it can cause tests to randomly fail, which is a problem.

The issue is the lack of synchronization between the stored waker and the state of the cursor.

There are a few possible fixes:

1. Use a single lock around the entire cursor state, instead of per-field locks.
2. Store the waker before performing any operation which might block, or at least re-check if the operation can succeed after storing the waker.
3. Look for a prebuilt solution, it seems like something like this should exist in futures or elsewhere.

[jbr]

A focused pr with any of these fixes would be welcome 👍. I didn't find anything preexisting, but it is a fairly general-purpose tool and ideally would not be defined or maintained in async-h1. Preference should be for 3, but since our usage is just for tests, either of the other two that works would be fine