For loops: Allow sharing variables with main program

[ajor]

This lets programs such as this run:

BEGIN
{
  @map[0] = 1; @map[1] = 2;
  $cnt = 0;
  for ($kv : @map) {
    $cnt++; // Previously would have errored with "Variables defined outside of a for-loop can not be accessed in the loop's scope"
  }
  print($cnt);
}

Implementation Details

1. Determine which variables need to be shared with the loop callback
2. Pack pointers to them into a context struct
3. Pass pointer to the context struct to the callback function
4. In the callback, override the shared variables so that they read and write through the context pointers instead of directly from their original addresses

Pseudo code for the transformation we apply:

Before:

PROBE {
  $str = "hello";
  $not_shared = 2;
  $len = 0;
  @map[11, 12] = "c";
  for ($kv : @map) {
    print($str);
    $len++;
  }
  print($len);
  print($not_shared);
}

After:

struct ctx_t {
  string *str;
  uint64 *len;
};
PROBE {
  $str = "hello";
  $not_shared = 2;
  $len = 0;
  @map[11, 12] = "c";

  ctx_t ctx { .str = &$str, .len = &$len };
  bpf_for_each_map_elem(@map, &map_for_each_cb, &ctx, 0);

  print($len);
  print($not_shared);
}
long map_for_each_cb(bpf_map *map,
                     const void *key,
                     void *value,
                     void *ctx) {
  $kv = (((uint64, uint64))key, (string)value);
  $str = ((ctx_t*)ctx)->str;
  $len = ((ctx_t*)ctx)->len;

  print($str);
  $len++;
}

Checklist

N/A: For-loops haven't been released yet and this is basic functionality expected of for-loops so is covered by the existing docs/changelog.
- [ ] Language changes are updated in man/adoc/bpftrace.adoc and if needed in docs/reference_guide.md
- [ ] User-visible and non-trivial changes updated in CHANGELOG.md

• The new behaviour is covered by tests

[jordalgo]

Nit: I know f is more terse but I much prefer the variable name for_loop.

[jordalgo]

naming nit: maybe make this consistent with the method e.g. anon_types_ or AddAnonymousStruct

[jordalgo]

Why make the value generic, it seems like all the places where this is used are creating a AllocaInst?

[jordalgo]

Since this is class level is there a case of possible collision where two (or more probes) use the same scratch variable name and reference them in a loop?

Or no, because each prog has it's own ELF file...

[jordalgo]

super duper nit

Suggested change

	for (size_t i = 0; i < ctx_fields.size(); i++) {
	for (size_t i = 0; i < ctx_fields.size(); ++i) {

[jordalgo]

super nit (makes it slightly easier to grok IMHO)

Suggested change

	auto *field_expr = variables_[field.name].value;
	auto *field_ty = field_expr->getType();
	auto *field_ty = variables_[field.name].value->getType();

[jordalgo]

I like the example!

[jordalgo]

Nit: looks like you could combine these two loops to get referenced and new vars.

[jordalgo]

Few small questions and nits but otherwise LGTM.