No longer let the GC replace dead keys with the LJ_TDEADKEY tag.

Important: this changes the semantics of the write barrier!
Carefully read the big comment block in lj_obj.h
This helps HREFK key slot specialization and allows safely hoisting
HREF/HREFK across GC steps, too (fix for a barely reproducible bug).
Dead keys are only removed during a table resize (as before).

src/lj_opt_fold.c

@@ -1351,8 +1351,9 @@ LJFOLDF(fwd_xload)
 /* Write barriers are amenable to CSE, but not across any incremental
 ** GC steps.
 **
-** The same logic applies to open upvalue references, because the stack
-** may be resized during a GC step.
+** The same logic applies to open upvalue references, because a stack
+** may be resized during a GC step (not the current stack, but maybe that
+** of a coroutine).
 */
 LJFOLD(TBAR any)
 LJFOLD(OBAR any any)