Update cast of load16

Bitwise operations promote the cast to uint16_t to int. So to be consistent, first cast to uint to keep unsignedness and prevent implicit promotion. Then cast back to uint16_t after all bitwise operations are completed. 

This prevents compiler warnings, unneccessary casts and in the case of right shifting (not the case here) unexpected behaviour.

sse/blake2-impl.h

@@ -72,8 +72,8 @@ static BLAKE2_INLINE uint16_t load16( const void *src )
  return w;
 #else
  const uint8_t *p = ( const uint8_t * )src;
- return (( uint16_t )( p[0] ) << 0) |
- (( uint16_t )( p[1] ) << 8) ;
+ return ( uint16_t )((( uint32_t )( p[0] ) << 0) |
+  (( uint32_t )( p[1] ) << 8));
 #endif
 }