test: refactor benchmark latency PRINT_F() macro

Reduces the stack usage. Also separates the formatting of both the
cycle and nsec printing by pre-printing them to string.

Signed-off-by: Peter Mitsis <peter.mitsis@intel.com>