Decorrelated the wait abort ISR test's interrupt source from the thread it samples

[fdesbiens]

Follows #644 and #649, which bounded this test and lowered what it asks for. This
addresses the cause they worked around.

What the resonance actually is

The test waits for the periodic timer interrupt to land while _tx_thread_preempt_disable
is set. The same interrupt also puts the semaphore that wakes the thread being sampled,
so the two run in lockstep: the tick wakes thread 0, thread 0 does a fixed amount of work
and suspends, and the next tick arrives a fixed interval later with thread 0 in the same
place every time.

The window is a few instructions wide, so a sampler locked to the sampled thread's own
cycle can miss it indefinitely. Perturbing the handler's duration, which the handler
already does after 100 misses, shifts the phase but does not decorrelate the two.

The fix

_tx_linux_timer_interrupt waits on _tx_linux_timer_semaphore with a one-tick deadline
and delivers an interrupt early when the semaphore is posted — a mechanism the port
already relies on in _tx_thread_schedule. The test now runs a plain POSIX thread that
posts it, injecting interrupts at moments unrelated to the tick grid.

It posts only when nothing is outstanding, so interrupts can never be queued faster than
they are serviced and the injector cannot starve the system. It is confined to the Linux
simulation port by #ifdef __linux__, and no port file changes — the semaphore is a
global with external linkage, declared the same way tx_thread_schedule.c declares it.

Measured

Six build configurations of a branch carrying this change:

	Windows reached	Time
Before	1, 1, 2, 3, 3, 3 of 3	43–181 s each, three pinned at the budget
After	10 of 10, all six	under 1 s each

The test job on that branch went from 20m48s to 4m27s.

The count goes back to ten on the same reasoning that lowered it to three — ask for what a
run can actually reach. The budget and the zero-window ceiling are untouched, so a run
that somehow still falls behind behaves exactly as it does today.

The SMP copy deliberately does not get this

#649 kept the two copies logically identical, and the budget and ceiling logic still is.
The injector is the exception, because only the tx copy has the problem it solves.

Measured on the SMP copy: 8.9 seconds with the injector against 0 to 1 without, for
the same twenty windows. It has never been in the slow mode, and the extra interrupt
traffic contends across the simulated cores. Applying it there would be a straight
regression, so its count stays at twenty and its file is unchanged.

Verification

• tx suite 105/105, the test at 0.44 s; four direct runs reach 10 of 10 in 0–1 s.
• SMP suite 117/117 unmodified, its own test 0–1 s over four runs.