K_SEM_DEFINE is not a static initializer (requires runtime k_sem_init)

[swoisz]

Boreas's K_SEM_DEFINE requires a runtime k_sem_init() call after the macro; upstream's produces a fully-initialized struct usable before main(). SYS_INIT-time and constructor-time consumers hit this as a silent (or crashing) ordering dependency.

Fix: a real compile-time initializer, the way K_TIMER_DEFINE already does it. Depends on whether StaticSemaphore_t can be made constructor-free (e.g. lazy xSemaphoreCreateCountingStatic on first use behind an init-flag, or init via a Boreas SYS_INIT at PRE_KERNEL priority as a stopgap).

[swoisz]

Completed across the conversion series: K_SEM_DEFINE became a true compile-time initializer in #41 (notification-backed k_sem — the FreeRTOS control block and its ctor machinery are gone), and K_EVENT_DEFINE likewise in #42. The remaining named item (constructor-time usability) is regression-tested (test_sem_define_usable_in_constructor).