Optimize memory usage on Mac by pooling memory passed to sysctl

[cmacknz]

The sysctl implementation allocates 256 KiB for every invocation.

In go-sysinfo a sync.Pool is used to minimize the amount of memory that is allocated for each call. It uses a buffer from the pool to make the call then allocates only the size required rather than ARG_MAX.

Port this same optimization to the system metrics package. Here is one example where it would apply:

elastic-agent-system-metrics/metric/system/process/process_darwin.go

Lines 154 to 161 in 5f48574

	func getProcArgs(pid int, filter func(string) bool) ([]string, string, mapstr.M, error) {
	mib := []C.int{C.CTL_KERN, C.KERN_PROCARGS2, C.int(pid)}
	argmax := uintptr(C.ARG_MAX)
	buf := make([]byte, argmax)
	err := sysctl(mib, &buf[0], &argmax, nil, 0)
	if err != nil {
	return nil, "", nil, fmt.Errorf("error in sysctl: %w", err)
	}

We could avoid allocating a new slice for every process with this optimization.

[fearful-symmetry]

Something else to consider: Instead of allocating a single, large chunk of memory ahead of time, we can use sysctl's own memory allocation reporting to make sure we only allocate what memory we need. For example, on MacOS 12, we will allocate ARG_MAX bytes, which is 1048576 but the average process will actually need something closer to only 600.