fix(serial): fix SIGSEGV and file descriptor leaks in attachment constructor

[Code-Hex]

Why

Following merged PR #217, NewFileHandleSerialPortAttachment crashed with SIGSEGV because an invalid read or write descriptor caused an autoreleased NSError to be used by Go after its Objective-C autorelease pool drained. Additionally, two duplicated file descriptors leaked because locally allocated NSFileHandle objects were not released after attachment initialization retained them. Finally, CI full VM tests were being cancelled at 3 and 5 minutes despite making progress.

Scope

• The Objective-C bridge now retains the NSError across the pool boundary; Go converts it to a plain Go error and releases the retained NSError.
• The bridge releases both NSFileHandles on success, and releases the first if the second duplication fails.
• Added tests in serial_console_test.go to cover invalid read and write handles yielding EBADF and no attachment, and to ensure absence of descriptor leaks on failure and after success release.
• Added a test helper to count /dev/fd names via open directory Readdirnames, avoiding os.ReadDir which caused fstatat bad file descriptor errors on macOS 15 Intel.
• Increased the CI test job timeout in .github/workflows/compile.yml from 3 to 10 minutes.

Blast Radius

Changes are scoped entirely to NewFileHandleSerialPortAttachment error handling, file descriptor lifecycle, serial console tests, and the CI compile workflow timeout.

Verification

• Local targeted make test/run TARGET=TestNewFileHandleSerialPortAttachment passed.
• GitHub Actions run https://github.com/Code-Hex/vz/actions/runs/36408696564 completed successfully: formatting, the macOS 15 Intel full test job, and every matrix build job passed.