Fix thread_profile_frames crashing due to uninitialized PC

ZJIT leaves cfp->pc uninitialized in many instances, and the conditions
in thread_profile_frames() was outdated, leading sampling profilers to
crash.

Check `cfp->pc` for validity in a way that works for both YJIT and ZJIT.

Author: XrXr

vm_backtrace.c

@@ -1729,7 +1729,6 @@ thread_profile_frames(rb_execution_context_t *ec, int start, int limit, VALUE *b
 {
     int i;
     const rb_control_frame_t *cfp = ec->cfp, *end_cfp = RUBY_VM_END_CONTROL_FRAME(ec);
-    const rb_control_frame_t *top = cfp;
     const rb_callable_method_entry_t *cme;
 
     // If this function is called inside a thread after thread creation, but
@@ -1761,11 +1760,9 @@ thread_profile_frames(rb_execution_context_t *ec, int start, int limit, VALUE *b
             }
 
             if (lines) {
-                // The topmost frame may not have an updated PC because the JIT
-                // may not have set one.  The JIT compiler will update the PC
-                // before entering a new function (so that `caller` will work),
-                // so only the topmost frame could possibly have an out of date PC
-                if (cfp == top && cfp->jit_return) {
+                VALUE *iseq_encoded = ISEQ_BODY(cfp->iseq)->iseq_encoded;
+                // The JIT may not set the PC, so a range check is necessary
+                if (cfp->pc < iseq_encoded || cfp->pc > iseq_encoded) {
                     lines[i] = 0;
                 }
                 else {