The argument to ulp_run should be expressed in 32-bit words. Both the
address of ulp_entry and RTC_SLOW_MEM already are uint32_t*, so their
difference is the difference in addresses divided by sizeof(uint32_t).
Therefore the extra division by sizeof(uint32_t) is not needed.
- call rtc_gpio_init for pins which aren't configured as RTC by default
- for RTC IOs 16 and 17, read higher part of the input register
- automatically convert GPIO number into RTC IO number
- use ulp_set_wakeup_period instead of writing to register
