Fix unaligned access to `double` in RFloat

internal/numeric.h

@@ -206,10 +206,22 @@ rb_float_flonum_value(VALUE v)
     return 0.0;
 }
 
+#if SIZEOF_VALUE >= SIZEOF_DOUBLE || defined(UNALIGNED_WORD_ACCESS)
+# define UNALIGNED_DOUBLE_ACCESS 1
+#else
+# define UNALIGNED_DOUBLE_ACCESS 0
+#endif
+
 static inline double
 rb_float_noflonum_value(VALUE v)
 {
+#if UNALIGNED_DOUBLE_ACCESS
     return RFLOAT(v)->float_value;
+#else
+    double d;
+    memcpy(&d, &RFLOAT(v)->float_value, sizeof(double));
+    return d;
+#endif
 }
 
 static inline double

numeric.c

@@ -951,7 +951,11 @@ rb_float_new_in_heap(double d)
 {
     NEWOBJ_OF(flt, struct RFloat, rb_cFloat, T_FLOAT | (RGENGC_WB_PROTECTED_FLOAT ? FL_WB_PROTECTED : 0));
 
+#if UNALIGNED_DOUBLE_ACCESS
     flt->float_value = d;
+#else
+    memcpy(&flt->float_value, &d, sizeof(double));
+#endif
     OBJ_FREEZE((VALUE)flt);
     return (VALUE)flt;
 }