auto-v_dont_optimize for vcl_init only functions does not quite work yet

[nigoroll]

The intention behind 8551b1f was to ensure that we add v_dont_optimize for subs called from housekeeping subs, but only if they are called from there only (in other words, if a sub is called both from housekeeping and a any other sub, we do want the compiler to optimize).
This does work for a case like

vcl 4.1;
backend foo None;
import directors;

sub a {
	new d = directors.shard();
}

sub vcl_init {
	call a;
}

because the new keyword makes the sub "init-only":

void v_dont_optimize v_matchproto_(vcl_func_f)
VGC_function_a(VRT_CTX, enum vcl_func_call_e call,
    enum vcl_func_fail_e *failp)
{
  assert(call == VSUB_STATIC);
  assert(failp == NULL);
  assert(ctx->method & (VCL_MET_INIT));
...

however, if a sub does not have new and is still called from vcl_init only, it will not get the attribute:

vcl 4.1;
backend foo None;

sub a { }

sub vcl_init {
	call a;
}

->

void v_matchproto_(vcl_func_f)
VGC_function_a(VRT_CTX, enum vcl_func_call_e call,
    enum vcl_func_fail_e *failp)
{
  assert(call == VSUB_STATIC);
  assert(failp == NULL);
  assert(ctx->method & (VCL_MET_BACKEND_ERROR|VCL_MET_BACKEND_FETCH|VCL_MET_BACKEND_RESPONSE|VCL_MET_DELIVER|VCL_MET_FINI|VCL_MET_HASH|VCL_MET_HIT|VCL_MET_INIT|VCL_MET_MISS|VCL_MET_PASS|VCL_MET_PIPE|VCL_MET_PURGE|VCL_MET_RECV|VCL_MET_SYNTH));

The reason is that the okmask allows it to be called from anywhere. But v_dont_optimize should not be determined by from where can be called, but rather by from where it actually is called.