Created a workaround for a bug in MSAN for va_arg(,double)

MDEV-22691 MSAN use-of-uninitialized-value in test maria.maria-recovery2

This caused all my_vsnprintf() using doubles to fail.
Thanks to the workaround, I was able to remove the disabling of
MSAN in dtoa().

strings/dtoa.c

@@ -2168,9 +2168,6 @@ static int quorem(Bigint *b, Bigint *S)
 
 static char *dtoa(double dd, int mode, int ndigits, int *decpt, int *sign,
                   char **rve, char *buf, size_t buf_size)
-#if __has_feature(memory_sanitizer)
-  __attribute__((no_sanitize("memory"))) // FIXME: dd is claimed uninitialized
-#endif
 {
   /*
     Arguments ndigits, decpt, sign are similar to those

strings/my_vsnprintf.c

@@ -704,7 +704,13 @@ size_t my_vsnprintf_ex(CHARSET_INFO *cs, char *to, size_t n,
     }
     else if (*fmt == 'f' || *fmt == 'g')
     {
+#if __has_feature(memory_sanitizer)         /* QQ: MSAN has double trouble? */
+      __msan_check_mem_is_initialized(ap, sizeof(double));
+#endif
       double d= va_arg(ap, double);
+#if __has_feature(memory_sanitizer)        /* QQ: MSAN has double trouble? */
+      __msan_unpoison(&d, sizeof(double));
+#endif
       to= process_dbl_arg(to, end, width, d, *fmt);
       continue;
     }