Optimize system_time function

The uint128 code does not optimize as well as expected. Since this
function can be called quite often, use a more optimal implementation
(quite similar to what we had for x86_64 before refactoring).

Unfortunately this is still not as good as the hand-optimized assembler
previously used for 32bit x86, unless compiled with clang, where it gets
pretty close.

src/system/boot/arch/x86/cpu.cpp

@@ -304,8 +304,10 @@ slower_sample:
 extern "C" bigtime_t
 system_time()
 {
-	uint128 tsc(rdtsc());
+	uint64 tsc = rdtsc();
-	return uint64((tsc * gTimeConversionFactor) >> 32);
+	uint64 lo = (uint32)tsc;
+	uint64 hi = tsc >> 32;
+	return ((lo * gTimeConversionFactor) >> 32) + hi * gTimeConversionFactor;
 }