Fix cache alignment issues for software interrupt tracking

Oh no! These are not cache aligned. This is a large amount of false sharing that will cause a big perf decrease on multicore.

We want to essentially have each atomic be 128 bytes apart (on 128-aligned boundaries). The typical way to accomplish this without blowing up the cache and memory consumption is to put the counters in a struct, and align the structs.

struct event_counts {
    _Atomic volatile sig_atomic_t software_interrupt_counts_sigalarm_reply;
    _Atomic volatile sig_atomic_t software_interrupt_counts_kernel;
    /* ... */     
} __attribute((aligned(128)));

struct event_counts software_interrupt_counts[runtime_worker_threads_count];

Locate the software_interrupt_counts entry point and the event_counts structure described in the issue. Verify how the counters are allocated and accessed, then confirm that each atomic counter is separated by 128 bytes on 128-aligned boundaries without changing software interrupt tracking behavior.