fix(core): stop reconnecting cancelled streams (#3882)

### Description

Cancelling a reconnecting framed stream could be mistaken for a clean, incomplete EOF. The pending pull then reopened the World stream after cancellation, leaving the new reader unowned; local World consumers retained emitter listeners and polling intervals, eventually producing `MaxListenersExceededWarning` after repeated reads of the same durable stream.

The reader now latches cancellation across pending reads, completion checks, and reconnect acquisition. Reconnect work stops after cancellation, and a World stream that finishes opening after cancellation is immediately cancelled instead of being installed as the active reader.

### How did you test your changes?

Added focused regressions for cancellation while completion metadata is pending and while a reconnect acquisition is pending. The complete `@workflow/core` suite passes: 107 test files passed, 1 skipped; 2,285 tests passed, 3 expected failures, and 1 skipped. `@workflow/core` also typechecks. Before applying the fix, a 20-turn local eve session deterministically retained one World reader per turn and warned on listener 11; with this change, the same run peaked at one reader and ended with zero.