Linux thread pinning: capture the original affinity mask before pinning the main thread

New threads inherit the creating thread's CPU affinity (sched_getaffinity / pthread_create). If a placement helper pins the main thread to one core first and only then starts worker threads, every worker starts with that single-core mask; validating each worker's requested mask against its own inherited mask then wrongly refuses (or, if you widen to make it pass, escapes the job's allocated cores). Read and freeze the process's original allowed mask (e.g. sched_getaffinity(0) at startup, or the taskset/cgroup lane) before any pinning, validate every role's mask as a subset of that frozen original, never widen beyond it, and record requested vs actual masks per thread (/proc/<pid>/task/<tid>/status Cpus_allowed_list). Treat a failed parse/apply/verify as a hard stop, not 'placement skipped'.