Fix crash on invalid subtype indication.

[NikLeberg]

Continuation of the fuzzing crashes found by @avelure.
See #1038 (comment).

In resolve_resolution we expect rname to be of kind T_REF or T_ELEM_RESOLUTION otherwise an assert triggers. I simply added a check for the kind and prevent calling resolve_resolution. If this erroreous input is now given, the parser no longer crashes but instead maneuvers completely off the line and everything afterwards may generate lots of other errors. For such a rare and mean input I think its ok though. What do you think?

Cheers

[nickg]

I think it would be better to adjust resolve_resolution instead so that it handles this case. Something like

switch (tree_kind(rname)) {
case T_ELEM_RESOLUTION:
   ...
case T_REF:
   ...
default:
   error_at(tree_loc(rname), "not a valid resolution function name");
   tree_set_type(rname, type_new(T_NONE));
   break;

[NikLeberg]

Noted and changed.
Thanks for the pointers. I can now finally understand what the heck that all actually was and why the error was triggered. I redid the test in a cleaner way.