Hi xmas79
1. Accuracy: If you have a large enough signal such as 0.22Vpeak, there is no need to set PGA of 2 to get higher accuracy. You can maintain PGA of 1 and still get the same level of accuracy. By this way, you need not worry about compromising on the attenuation resistor value.
2. The measurement registers will have some offset error in them due to inherent noise in the system. Offset error is the measurement obtained when you apply zero input. The ADC offset error (refer specifications table and terminology section of datasheet) also contributes to this. So, it is fine if you notice a non-zero value when you read VRMS or IRMS register. By using the VRMSOS and IRMSOS registers, you can offset-compensate the RMS measurements, thus bringing this value closer to zero. After compensation, you can expect to meet the accuracy specified in the specifications table. IRMS is specified for less than 0.5% error till FS/100 while VRMS is specified only till FS/20 for the same amount of error. So, getting an accurate RMS register result at close-to-zero input conditions is not practical. For calibration, I would use an input signal which is more accurate than the expected measurement quantity's accuracy.
3. When the input signal is less than FS/10000, the IC's no-load feature kicks in. This feature stops the energy measurement from increasing by very small quantities due to noise. So, if you read energy registers, you will not see this happen. Can you try it out? If you still notice the issue, can you report the values you read from registers over a period of time?
Regards
hmani