STM32F3xx series also needs a wait of at least one clock cycle before
reading the BSY bit during a flash erase - previously this was only
applied to STM32F1xx series.
While not waiting for the BSY flag to clear works on STM32F103C8T6, it
does not on APM32F103C8T6. Only one half-word gets written while the
other one gets lost.
