Add tm_cause_interrupt_sync() for in-line ISR

[jserv]

interrupt_processing was measuring the same trap-plus-ISR-thread path interrupt_preemption_processing already exercises. Introduce tm_cause_interrupt_sync() so that test measures only the cost of the ISR body plus the semaphore handshake, and route interrupt_processing through the new entry.

Each port supplies both primitives:
tm_cause_interrupt() trap/IRQ-pend/ISR-thread path (unchanged)
tm_cause_interrupt_sync() inline tm_interrupt_handler() on the
caller's stack

FreeRTOS Cortex-M brackets the inline call with cpsid i/cpsie i and sets a benchmark-only flag so tm_{thread_resume,semaphore_put} take the FromISR path during the handler -- functionally safe under PRIMASK, no real ISR or context switch can race. ThreadX Cortex-M emulates an ISR preamble (system_state++ / preempt_disable++) under cpsid i and defers any handler-driven preemption to tm_threadx_benchmark_sync_complete() after cpsie i, mirroring real ISR semantics. POSIX hosts call the handler directly: their semaphore and resume APIs are mutex-guarded and safe from thread context.

---

Summary by cubic

Add tm_cause_interrupt_sync() to run the interrupt handler in-line and remove trap/ISR-thread overhead from interrupt_processing. This separates the two interrupt benchmarks and improves measurement accuracy.

• New Features

  • Added tm_cause_interrupt_sync() in tm_api.h and documented it in README; tm_cause_interrupt() remains for the real IRQ/trap path.
  • Updated interrupt_processing.c to use tm_cause_interrupt_sync() so it measures only the ISR body plus the semaphore handshake.
  • Port updates: FreeRTOS Cortex‑M masks IRQs and uses a benchmark flag so tm_thread_resume/tm_semaphore_put take the FromISR path safely; ThreadX Cortex‑M emulates ISR preamble and defers preemption via tm_threadx_benchmark_sync_complete(); POSIX calls the handler directly.

• Migration

  • Ports must implement both tm_cause_interrupt() and tm_cause_interrupt_sync(); if no safe inline path exists, alias the sync variant to the async one.

^{Written for commit 2696506. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 7 files

[github-actions]

Thread-Metric Benchmark Results

✅ All tests passed — 425e648 — 2026-05-13T05:05:51Z

Summary

RTOS	Target	OS	Status	Passed	Failed	Total
freertos	cortex-m	ubuntu-24.04	✅ passed	8	0	8
freertos	posix	macos-latest	✅ passed	8	0	8
freertos	posix	ubuntu-24.04	✅ passed	8	0	8
threadx	cortex-m	ubuntu-24.04	✅ passed	8	0	8
threadx	posix	macos-latest	✅ passed	8	0	8
threadx	posix	ubuntu-24.04	✅ passed	8	0	8

Per-Test Results

Test	freertos/cortex-m	freertos/posix	freertos/posix	threadx/cortex-m	threadx/posix	threadx/posix
basic_processing	✅	✅	✅	✅	✅	✅
cooperative_scheduling	✅	✅	✅	✅	✅	✅
preemptive_scheduling	✅	✅	✅	✅	✅	✅
message_processing	✅	✅	✅	✅	✅	✅
synchronization_processing	✅	✅	✅	✅	✅	✅
memory_allocation	✅	✅	✅	✅	✅	✅
interrupt_processing	✅	✅	✅	✅	✅	✅
interrupt_preemption_processing	✅	✅	✅	✅	✅	✅