fix(coro): propagate targeted spawn affinity through task chains

[Coldwings]

Description

Fix scheduler::go_to() and scheduler::go_joinable_to() so their documented worker affinity remains continuous across the directly awaited Elio task chain that implements the spawned vthread.

Previously, targeted spawn set affinity only on the scheduler-owned callable wrapper. The task returned by the callable had a separate execution context and therefore resumed with NO_AFFINITY after its first non-I/O suspension. Subsequent I/O could then bind to a different worker's io_context.

Direct Elio task awaits now inherit caller-requested affinity when the child has none. When the child finishes, its final caller-requested affinity is copied back to its Elio awaiter, so an explicit clear_affinity() or set_affinity() in a deeper frame remains authoritative for the rest of the same logical vthread. I/O ownership remains operation-local and continues to take precedence through effective_affinity().

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Performance improvement (optimization that improves speed/memory usage)
• Documentation (changes to documentation, comments, or examples)
• Refactoring (code changes that neither fix bugs nor add features)
• Tests (adding or modifying tests)
• Build/CI (changes to build system, CI configuration, or dependencies)

Related Issues

Closes #1006
Related to #1003

Changes Made

Core Changes

• Propagate user affinity from an Elio awaiter into a directly awaited task<T> or task<void> when the child has no affinity.
• Propagate the child's final user affinity back to the awaiter before returning a value or rethrowing an exception.
• Preserve the existing symmetric-transfer and frame-ownership model; the fix does not enqueue borrowed child handles.
• Remove the scheduler-wrapper-only propagation helper because task-level propagation now covers the complete directly awaited chain.
• Add deterministic regressions for go_to() and go_joinable_to() through task_scope(), a value-returning channel::recv() task, a cross-worker wake, and subsequent real socket I/O.
• Verify inherited affinity, clear_affinity(), and set_affinity(1, false) behavior.

API Changes (if applicable)

None. This restores the existing documented vthread and targeted-spawn behavior.

Migration Guide (if breaking change)

Not applicable.

Testing

Unit Tests

• Added new tests for the changes
• Updated existing tests if needed
• All tests pass locally

The regression waits on a bounded channel and uses the channel's waiter-publication test hook. It then runs a marker on worker 0 before waking from worker 1, proving that the original await_suspend() has returned before cross-worker resumption. One normal scheduler insertion is deliberately rejected so the test distinguishes affinity routing from local fallback. After the nested task chain returns, a cancellable socket receive verifies the owning io_context, completion worker, and post-I/O user affinity.

Integration Tests

• Tested with existing examples
• Tested in real-world scenarios (if applicable)

Sanitizer Testing

• Tested with ASAN (AddressSanitizer)
• Tested with TSAN (ThreadSanitizer)
• No new warnings or errors

Test Results

Local compilation and tests were not run; executable validation is delegated to the GitHub Actions matrix.
Static validation: git diff --check passed.
Independent read-only review: no unresolved actionable findings on the pushed design.
GitHub Actions at exact head 3514da6: all 10 checks passed, including x64/arm64 Debug sanitizer jobs.

Checklist

Code Quality

• My code follows the project's code style
• I have added/updated comments for complex logic where needed
• I have removed any debug code, TODOs, or commented-out code
• My changes generate no new warnings

Documentation

• I have updated documentation (not needed; the public contract is unchanged)
• I have added examples for new features (not applicable)
• I have updated API documentation (not applicable)

Testing

• I have added tests that prove my fix is effective or my feature works
• New and existing unit tests pass locally with my changes
• I have tested with ASAN and TSAN

Compatibility

• My changes are backward compatible
• I have considered the impact on existing users
• I have updated CHANGELOG.md (not applicable; no API change)

Performance (if applicable)

• I have considered the performance impact
• I have added benchmarks for performance-critical changes (not applicable)

Screenshots / Diagrams

Not applicable.

Additional Notes

The fix intentionally does not alias execution contexts across coroutine frames. Cancellation continues to use the existing one-way direct-await link, while operation-local I/O pins remain owned by the frame and awaitable that registered the operation. Only caller-requested affinity is carried across the logical vthread chain.

The first regression implementation used a custom raw-handle gate. TSAN correctly identified that the test could resume and destroy the frame while its await_suspend() was still accessing the gate. The final test uses the channel publication hook plus a same-worker marker to establish the required happens-before ordering.

Reviewer Guidance

Areas requiring special attention:

• Whether direct Elio task awaits are the correct boundary for vthread affinity continuity.
• Whether downward inheritance and upward propagation preserve explicit clear_affinity() / set_affinity() behavior without copying I/O ownership.
• Whether the bounded-channel publication and marker sequence excludes concurrent destruction of an active await_suspend() frame.
• Whether every test failure path shuts down the scheduler before borrowed channel, observation, cancellation, and socket state is destroyed.

Questions for reviewers:

• Are there any direct Elio task-await boundaries that should intentionally terminate user-affinity propagation, analogous to the existing foreign-coroutine cancellation boundary?

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Coldwings]

Independent read-only review completed against exact head 3514da6.

Review scope covered direct-Elio-await affinity propagation, symmetric-transfer/frame ownership, cancellation and exception paths, explicit affinity overrides, I/O-pin precedence, scheduler shutdown, and the sanitizer safety of the regression. No unresolved actionable findings remain.

Two issues found during earlier review iterations were corrected before this head: the fix no longer asynchronously queues a borrowed directly-awaited child handle, and the regression now uses bounded-channel waiter publication plus a same-worker marker instead of a raw-handle gate. The final exact-head review found no contract, lifetime, sanitizer, compile, hang, teardown, or portability blockers. GitHub Actions Debug sanitizer jobs are still running.