AP_HAL: prevent a single bad transfer causing an IMU to be marked unhealthy

This is a response to this issue:
https://discuss.ardupilot.org/t/gyro-problem-after-waypoint-log-analisys

The 2nd gyro went bad with a large offset. Then the first gyro was
marked unhealthy, forcing the EKF to switch to the 2nd gyro. That
resulted in a crash.

I think the SPI bus was getting bad transfers and the register check
code happened to get a bad transfer, thereby marking the first gyro
unhealthy

This change ensures we only fail the register check if two transfers
in a row are bad. This makes it much less likely that a noisy bus will
lead to an unhealthy gyro

libraries/AP_HAL/Device.cpp

@@ -99,7 +99,7 @@ bool AP_HAL::Device::check_next_register(void)
     _checked.counter = 0;
 
     struct checkreg &reg = _checked.regs[_checked.next];
-    uint8_t v;
+    uint8_t v, v2;
 
     if (_bank_select) {
         if (!_bank_select(reg.bank)) {
@@ -113,7 +113,8 @@ bool AP_HAL::Device::check_next_register(void)
         }
     }
 
-    if (!read_registers(reg.regnum, &v, 1) || v != reg.value) {
+    if ((!read_registers(reg.regnum, &v, 1) || v != reg.value) &&
+        (!read_registers(reg.regnum, &v2, 1) || v2 != reg.value)) {
         // a register has changed value unexpectedly. Try changing it back
         // and re-check it next time
 #if 0