Initial implementation of RTCMEM storage

[mcspr]

Part of #1265

Store system crash counter and reset reason in rtcmem instead of eeprom
Store relay state mask in rtc in addition to the eeprom
Store relay state in eeprom only when boot mode requires it
Simplify relay state mask calculation / reading via std::bitset

Missing:
energy counter (EEPROM_ENERGY_COUNT) - should it be saved like this?
mqtt json id counter (EEPROM_MESSAGE_ID) - can write here every send, no need for delay like for eeprom.

Note:
This also can directly use rtcmem as an array (example from esp-open-rtos)

typedef volatile uint32_t rtcmem_array_t[96];
#define RTCMEM_USER (*(rtcmem_array_t *)(RTCMEM_ADDR))
...
RTCMEM_USER[...SOME_INDEX_DEFINE...] = 1; // indexes defined like for eeprom
RTCMEM_USER[97] = 2; // compiler will yell because of rtcmem_array_t definition

And idk if consistency check is required at all, and just checking magic value would suffice
+ tracking resetInfo->reason for REASON_DEFAULT_RST / REASON_SOFT_RESTART

[xoseperez]

We are not checking for consistency in the flash atm.
I would add energy count too and light colors also (same update frequency than relays).

[mcspr]

for energy counter the goal here is to protect readings in between save intervals from disappearing?

lights are straightforward enough, save 5 channels + brightness + mireds state. even fits inside 64 bit. only unlike relay it always saves - should it have an option like relayBoot to optionally do it?

[mcspr]

Removed crc, added mqtt counter, added general methods for energy total and lights handling.
Last thing - check when it is appropriate to save / read them.
edit: and another - use module prefix instead of _rtcmem... for methods?

For lights: LIGHT_SAVE_ENABLED should be a setting and somehow wired into this?
For sensor: one can also save uptime() in rtc and reference that in energy ts code, since now it becomes 2 states - uptime under power with occasional reboots and complete shutdown with eeprom saves.

[xoseperez]

LGTM, but I would merge it after 1.13.4 so we have time to test it.

[mcspr]

Light issue resolved via static check of array size.

To reiterate - breaking changes are:
mqtt counter (no longer persisted between power-offs)
emon sensor total persists between reboots

I wonder if ntp time can be saved too.

[mcspr]

No obvious things stand up. Merging in the interest of moving milestone forward.
edit: ... perhaps "4 byte blocks" should be named word (or simply 32 bit) access. something to signify that we cannot read less than that. also, alignment...

Exception system might need some tweaking though:

• maybe lock the boards in safe mode until reboot. don't reset system stability counter to 0 on the 5th time, and force manual intervention (terminal command / web switch / power cycle)
• allow to store "small" crash log, instead of the full one