ZJIT: Avoid NoSingletonClass patchpoint when already a singleton class

The true purpose of assume_no_singleton_classes() is to retain soundness
in event of the effective class of values changing. When the effective
class is already a singleton class that can never happen. (Except due to
IO#reopen, but that's out of scope for now.)

This deletes the patchpoint for calls on `main` that happens a lot in
tests.

Author: XrXr

zjit/src/cruby.rs

@@ -492,6 +492,11 @@ impl VALUE {
         }
     }
 
+    pub fn is_singleton_class(self) -> bool {
+        // TODO(alan): clean up one of double check on T_CLASS
+        unsafe { RB_TYPE_P(self, RUBY_T_CLASS) && rb_zjit_singleton_class_p(self) }
+    }
+
     /// Return true for a static (non-heap) Ruby symbol (RB_STATIC_SYM_P)
     pub fn static_sym_p(self) -> bool {
         let VALUE(cval) = self;

zjit/src/hir.rs

@@ -2511,6 +2511,11 @@ impl Function {
             // This class can never have a singleton class, so no patchpoint needed.
             return true;
         }
+        if klass.is_singleton_class() {
+            // When a value has a singleton class, its effective class can't change anymore.
+            // No patchpoint needed.
+            return true;
+        }
         if has_singleton_class_of(klass) {
             // We've seen a singleton class for this klass. Disable the optimization
             // to avoid an invalidation loop.