RTC probe: field checks skip real RTCs, and only the RV3028 has an identity mask #38

rtc_probe() (dev f9d01cf) decides whether the device at a declared address is the RTC it expects. Three of its checks can skip a real RTC, and only one chip type is checked for the bits that would rule a device out. MeshCore's equivalent work, meshcore-dev/MeshCore#3544 (issue meshcore-dev/MeshCore#3546), grounds every rule in the vendor documents. This issue applies the same rules to ZephCore.

What is wrong

• Year: bcd_field_ok(yb, 99) skips a chip whose year byte is not valid BCD while its power-loss flag is clear. ZephCore writes y % 100, but a chip that ran MeshCore first can hold a year byte of A0 or above (RTClib writes bin2bcd(year - 2000)), and a real RTC is then never adopted.
• Hours: bcd_field_ok(hb, 23) after masking 0x3F fails for a DS3231 or RV3028 left in 12-hour mode by other firmware: a PM hour reads with bit 5 set, for example 0x32. Related to RTC probe assumes 24-hour mode: an RV3028 left in 12-hour mode is misread or never adopted #34.
• Always-zero masks: only an RV3028 descriptor is checked (rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35). The DS3231 and RX8130CE masks are documented as well; the PCF8563's are not, since its unused bits are "x = not relevant" (NXP PCF8563 data sheet Rev. 11.1, Table 4, p. 10).

Proposed rules (as meshcore-dev/MeshCore#3544)

1. A per-descriptor always-zero mask for the seven time registers:
   • DS3231 {80,80,80,F8,C0,60,00}: Maxim DS3231 19-5170 Rev. 10, Figure 1, p. 11.
   • RV3028 {80,80,C0,F8,C0,E0,00}: RV-3028-C7 Application Manual Rev. 1.4, §3.2, p. 12.
   • RX8130CE {80,80,C0,80,C0,E0,00}: Epson ETM50E-10, §13.2.1, Table 12, p. 22.
   • PCF8563: none.
2. All seven bytes reading 0xFF (an erased EEPROM), as today.
3. Seconds, minutes, date and month valid BCD in range, unless the power-loss flag is set. Never check the year or the hours.
4. A device is ruled out only on two reads that agree; a failed read never rules it out.

#35 refuses an RV3028 only when the same register shows an always-zero bit on both reads. meshcore-dev/MeshCore#3544 accepts two reads that rule the device out for different reasons. Both need two consecutive corrupted reads to misfire; this issue should state which it uses.

Hardware to test on

The second RAK4631 on this host carries two extra I2C modules, including an LPS22HB at 0x5C and a Bosch environment sensor at 0x76. meshcore-dev/MeshCore#3544's read-only test there found the LPS22HB passes the DS3231 rule (its WHO_AM_I value sits where OSF lives), which is a limit of flag-based identity.

Refs meshcore-dev/MeshCore#3544, #34, #35