Fix I2C address claiming and probe both BMP/BME addresses

[agessaman]

Bosch BMP/BME parts pick their I2C address from the SDO pin: 0x76 or 0x77. SENSOR_TABLE only ever had the 0x76 entry, so anything strapped high got seen by the bus scan and then quietly skipped. That covers Grove modules and most of the cheap BMP280 breakouts. A user hit this and found their sensor only worked once they moved it to 0x76.

There's already a per-variant TELEM_*_ADDRESS override, but it doesn't help here.

First commit claims an address once a driver has successfully initialized it. That's mostly groundwork for the second commit, but it fixes something on its own: several table entries share an address, and not every driver bothers to check a chip ID. INA226::begin() only checks that the address ACKs. In the default sensor_base both SHT4X and INA226 sit at 0x44, so a real SHT4x was also being registered as an INA226 and reporting junk current on a second channel.

Second commit adds the alternate-address entry for each Bosch sensor. There's one static driver instance per sensor type, so an entry whose query function is already active gets skipped—the second address is a fallback, not a second device. Without that check, two same-type parts at 0x76 and 0x77 would register two channels that both read whichever one initialized last.

On safety: entries whose address didn't ACK during scanI2CBus still never get touched, and all four drivers verify a chip ID before they'll claim an address (Adafruit_BMP280 against the chipid argument, Adafruit_BME280 against 0x60, bme68x_init against BME68X_CHIP_ID, Adafruit_BMP085 against 0x55). BMP085 stays last in the table, so a real BMP180 at 0x77 still ends up with its address after the newer drivers turn it down.

One flag: bme68x_init() does a soft reset before it reads the chip ID, so with BME680 compiled in, anything that ACKs at 0x77 now takes a blind 0xE0 <- 0xB6. That's the standard soft-reset register on Bosch parts, and it was already happening at 0x76, so I don't think it's a problem—but it is new behavior at the second address.

Built Heltec_v3_sensor, T_Beam_S3_Supreme_SX1262_repeater and LilyGo_T-Echo_repeater. The last two pin TELEM_BME280_ADDRESS=0x77, so they reverse the direction. The overrides are redundant now, but I left them alone rather than grow the diff. The fix was tested on a Station G3 using a BMP280 connected via Grove.