Make mod_p{224,256,384] a bit faster

Speedup is roughly 25%, giving a 6% speedup on ecp_mul() for these curves.
diff --git a/include/polarssl/bignum.h b/include/polarssl/bignum.h
index 769e546..eae15e0 100644
--- a/include/polarssl/bignum.h
+++ b/include/polarssl/bignum.h
@@ -128,6 +128,7 @@
 #define POLARSSL_HAVE_UDBL
 #else
   #if ( defined(_MSC_VER) && defined(_M_AMD64) )
+    #define POLARSSL_HAVE_INT64
     typedef  int64_t t_sint;
     typedef uint64_t t_uint;
   #else
@@ -137,6 +138,7 @@
           defined(__ia64__)  || defined(__alpha__)     || \
           (defined(__sparc__) && defined(__arch64__))  || \
           defined(__s390x__) ) )
+       #define POLARSSL_HAVE_INT64
        typedef  int64_t t_sint;
        typedef uint64_t t_uint;
        typedef unsigned int t_udbl __attribute__((mode(TI)));