Survive continuations posted after NetMQPoller disposal

[follesoe]

Fixes #9

Problem

NetMQPoller installs NetMQSynchronizationContext on the poller thread, so awaiting code captures it and may post continuations after the poller is disposed. Post then threw ObjectDisposedException out of Task.Start, which the await machinery rethrows on a thread-pool thread — uncatchable by application code and fatal to the process (Sentry BLUEYE-APP-FH, and BLUEYE-APP-9G as the native SIGABRT on iOS). Separately, tasks scheduled between Stop() and Dispose() were discarded with the task queue, leaving their awaiters hanging forever.

Changes

• NetMQSynchronizationContext.Post: falls back to ThreadPool.QueueUserWorkItem when the poller is disposed, including when disposal races the Task.Start call (the faulted task's exception is observed so it cannot surface as UnobservedTaskException).
• NetMQSynchronizationContext.Send: executes the callback inline on the calling thread in the same situations.
• NetMQPoller.Dispose: captures tasks still in m_tasksQueue (by enumerating the backing queue — TryDequeue depends on pair-socket signalling that is unavailable once the poller thread has exited) and executes them in order on a thread-pool thread, rather than discarding them. They are not run on the disposing thread, so arbitrary continuations cannot block or deadlock the caller of Dispose().

Behavioural note: a fallback callback runs without the poller's thread affinity. At that point the poller is already disposed, so the affinity guarantee no longer protects anything — the alternative was a process crash or a hung awaiter, not a working continuation.

Tests

Three new tests in NetMQPollerTest covering Post-after-dispose, Send-after-dispose, and task-scheduled-after-Stop. Full Poller category passes (26/26); the library builds clean for net10.0, net10.0-android36.0 and net10.0-ios26.0.

Upstreaming

After a BlueyeApp release confirms BLUEYE-APP-FH/9G stop occurring, this is a candidate to file upstream (closest upstream report is zeromq#697, stale-closed without a fix).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Prevents fatal crashes and hung awaiters caused by NetMQSynchronizationContext continuations being posted after a NetMQPoller has stopped/disposed, by providing safe fallback execution paths and ensuring queued tasks aren’t silently dropped during disposal.

Changes:

• NetMQSynchronizationContext.Post falls back to ThreadPool when the poller is disposed (including dispose races during Task.Start), avoiding uncatchable exceptions in the await continuation path.
• NetMQSynchronizationContext.Send falls back to inline execution on the calling thread in the same disposed/race scenarios.
• NetMQPoller.Dispose snapshots any remaining queued tasks and schedules them to run (in order) on the thread pool instead of discarding them.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/NetMQ/NetMQSynchronizationContext.cs	Adds disposed/race handling to Post/Send to avoid throwing and to ensure continuations still run.
src/NetMQ/NetMQPoller.cs	Drains and executes tasks left in m_tasksQueue during disposal so Stop→Dispose doesn’t strand awaiters.
src/NetMQ.Tests/NetMQPollerTest.cs	Adds regression tests for Post-after-dispose, Send-after-dispose, and tasks scheduled after Stop.
.github/workflows/CI.yml	Ensures CI tests run with net10.0 to match the build configuration.

---

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