benchmark/latency: support for Cortex-M targets

Remove x86 specific includes. Add Cortex-M specific instruction pipeline
flush helper.  Microkernel benchmark has been verified to work on STM32
MINI A15 (STM32F103VET) board. Due to memory constraints, Nucleo-F103RB
can only run a nanokernel version of the benchmark.

Change-Id: I2e49c240d6985ceb5643551397d6e0a8cc65b3c6
Signed-off-by: Maciek Borzecki <maciek.borzecki@gmail.com>