Enable ed25519-donna's SSE2 code when possible for 32 bit x86.

This probably requires the user to manually set CFLAGS, but should
result in a net gain on 32 bit x86. Enabling SSE2 support would be
possible on x86_64, but will result in slower performance.

Implements feature #16535.

changes/feature16535

+  o Minor features (performance)
+    - Improve the runtime speed of Ed25519 operations and Curve25519 keypair
+      generation when built targeting 32 bit x86 platforms with SSE2
+      available. Implements ticket 16535.

src/ext/ed25519/donna/README.tor

@@ -40,3 +40,7 @@ as of 8757bd4cd209cb032853ece0ce413f122eef212c.
  * On non-x86 targets, GCC's Stack Protector dislikes variables that have
    alignment constraints greater than that of other primitive types.
    The `ALIGN` macro is thus no-oped for all non-SSE2 builds.
+
+ * On 32 bit x86 targets that the compiler thinks supports SSE2, always
+   enable SSE2 support by force defining ED25519_SSE2 (x86_64 would also
+   always support this, but that code path is slower).

src/ext/ed25519/donna/ed25519-donna-portable.h

@@ -158,6 +158,16 @@ static inline void U64TO8_LE(unsigned char *p, const uint64_t v) {
 	#define ALIGN(x)
 #endif
 
+/* Tor: Force enable SSE2 on 32 bit x86 systems if the compile target
+ * architecture supports it.  This is not done on x86-64 as the non-SSE2
+ * code benchmarks better, at least on Haswell.
+ */
+#if defined(__SSE2__) && !defined(CPU_X86_64)
+	/* undef in case it's manually specified... */
+	#undef ED25519_SSE2
+	#define ED25519_SSE2
+#endif
+
 #include <stdlib.h>
 #include <string.h>