[#1898] Add test coverage for signal-isolation behavior

[zacheryasc]

The purpose is to complete test coverage that asserts the behavior #1898 prescribes, so a future fix PR can be validated against it. This PR does not fix #1898.

Coverage is spread across three test files:

• iceoryx2-bb/linux/tests-common/src/epoll_tests.rs — a signal subscribed via EpollBuilder::handle_signal is delivered as EpollEvent::Signal despite the global handler claiming the disposition
• iceoryx2-bb/linux/tests-common/src/signal_fd_tests.rs — signalfd delivery
• iceoryx2-bb/posix/tests-common/src/signal_tests.rs — call_and_fetch per-thread isolation, and the signal_test::signal_call_and_fetch_with_registered_handler_works() deadlocks in single-threaded execution #1458 contention case

Notes for Reviewer

• This is coverage prep for Signal handling lacks ownership and isolation across consumers, causing races and contention #1898, not the fix. The ignored tests fail today by
  design; a fix PR un-ignores them as proof.
• With the ignore removed, the epoll test was confirmed to fail deterministically
  in ~0.2 s (no hang, no watchdog trip).

Pre-Review Checklist for the PR Author

• Add sensible notes for the reviewer
• PR title is short, expressive and meaningful
• Consider switching the PR to a draft (Convert to draft)
  • as draft PR, the CI will be skipped for pushes
• Relevant issues are linked in the References section
• Branch follows the naming format (iox2-123-introduce-posix-ipc-example)
• Commits messages are according to this guideline
  • Commit messages have the issue ID ([#123] Add posix ipc example)
  • Keep in mind to use the same email that was used to sign the Eclipse Contributor Agreement
• Tests follow the best practice for testing
• Changelog updated in the unreleased section including API breaking changes

PR Reviewer Reminders

• Commits are properly organized and messages are according to the guideline
• Unit tests have been written for new behavior
• Public API is documented
• PR title describes the changes

References

Relates to #1898 #1458

[zacheryasc]

Reasoning for the change in the epoll test behavior:

• The previous watchdog-based guard would hang for over 10 minutes, and possibly flake if a signal slipped through somewhere, without actual guarantees on proper delivery.
• The rewrite deterministically succeeds or fails with a send-receive check that asserts proper epoll delivery.

[zacheryasc]

Reasoning for the change in the signalfd test behavior:

• The previous delivery test used an unbounded retry loop that would spin forever when the fd was starved, never actually failing. The rewrite deterministically succeeds or fails within a bounded ~200 ms poll.
• Two regression guards are added to pin the isolation contract any Signal handling lacks ownership and isolation across consumers, causing races and contention #1898 fix must preserve: dropping a SignalFd must not leave its signal monopolized away from the global handler, and owning one signal must not suppress delivery of an unrelated signal.

[zacheryasc]

Final coverage implemented. The reasoning for the change in signal test behavior:

• directly and deterministically show we do not contend for the same signal
• and we do not consume unrelated signals
• and we do not deadlock

[zacheryasc]

@elBoberido This should provide coverage for the general problem exposed from our previous conversation. This PR and the associated issue is ready for review.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 76.30%. Comparing base (e1fbd77) to head (f6caab6).
⚠️ Report is 23 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1900      +/-   ##
==========================================
+ Coverage   76.21%   76.30%   +0.09%     
==========================================
  Files         452      454       +2     
  Lines       45109    45404     +295     
  Branches     1436     1436              
==========================================
+ Hits        34378    34645     +267     
- Misses       9525     9553      +28     
  Partials     1206     1206              

Flag	Coverage Δ
CPP	62.71% <ø> (ø)
Rust	76.15% <ø> (+0.09%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 10 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[zacheryasc]

@elBoberido ping. This adds deterministic test coverage for race behavior and more when dealing with posix signals. It makes some extant tests more clear and straightforward, and checks for more possible violations of correct behavior.

[elBoberido]

@zacheryasc sorry, this totally slipped my attention. Maybe @elfenpiff can take a look at the PR. He is deeper into this part of the code. I once had a brief look at the code in signal.rs and came to the conclusion that one could remove quite some code and potentially also get rid of LAST_SIGNAL, which opens the door to not have to use the mutex and run into deadlocks in a single threaded signal execution.

[zacheryasc]

@zacheryasc sorry, this totally slipped my attention. Maybe @elfenpiff can take a look at the PR. He is deeper into this part of the code. I once had a brief look at the code in signal.rs and came to the conclusion that one could remove quite some code and potentially also get rid of LAST_SIGNAL, which opens the door to not have to use the mutex and run into deadlocks in a single threaded signal execution.

The issue, speaking broadly, is that posix signals are just very uncomfortable for concurrency, and twist your arm into building a singleton-router-ish signal handler. So we might solve single thread in such a way, but mutlithread is a different beast. I didn't make any architectural decisions here, though, except implied concurrency via the test construction. This is just setting guardrails to make correctness more explicit.