Stopped the Linux port losing a mutex wake-up to a suspend signal

[fdesbiens]

Part of #752. This is the ThreadX half of what #753 fixes for ThreadX SMP.

The Linux port has both halves of the same defect. Its suspend handler in ports/linux/gnu/src/tx_initialize_low_level.c calls sigsuspend and does not return until the thread is resumed, and tx_linux_mutex_lock in ports/linux/gnu/inc/tx_port.h is a bare pthread_mutex_lock. A thread can therefore be signalled while it is parked on _tx_linux_mutex; glibc waits for a contended mutex in a loop that re-arms the futex wait after a signal, and this handler never returns to it, so the next unlock hands its wake-up to a thread that will not act on it and any other waiter is left on a mutex that reads free. #752 has the captures of that state from the SMP port.

tx_linux_mutex_lock now calls a helper that waits with pthread_mutex_timedlock and retries, so the wait is re-armed every TX_LINUX_MUTEX_RETRY_NSEC — one millisecond — and a lost wake-up costs a retry rather than the process. Nothing else changes; the hand-over path is untouched whenever the wake-up arrives. pthread_mutex_timedlock needs _GNU_SOURCE under -std=c99, and ports/linux/gnu/CMakeLists.txt and the port's example Makefile both already define it.

These are the only two ports affected. A sigsuspend-based suspend handler exists in ports/linux/gnu and ports_smp/linux/gnu and nowhere else in the tree, and those two are also the only ports that take a critical section with pthread_mutex_lock. The Win32 ports use Windows primitives and are unaffected.

What is different from #753, and worth being explicit about: this port has never been observed to deadlock. The ThreadX regression suite recorded no first-attempt failure across 105 configuration-runs with retries disabled, and none in any of the 33 CI runs dispatched while the SMP defect was being characterised — including the two whose SMP job failed. That is consistent with one emulated core and far less suspend and resume traffic, not with the code being different. So this is a fix by inspection rather than one driven by a reproduction, and it is offered as such.

What it is verified to do is break nothing: all seven build configurations pass with retries disabled and CTEST_PARALLEL_LEVEL=1, 105/105 in five and 100/100 in two, 715 test executions.