C2 is probably useful. Your voltage divider has a relative high output impedance, and this reduces the accuracy of the ADC. A capacitor on the ADC input that is many times bigger then the capacitance of the ADC input itself provides a low impedance to the mostly capacitive input of the ADC. One of the uC manufacturers made an application note for this. (I think it was Atmel, but I'm not sure).
1.7 V RMS may be just a little too much for the 3.3 V supply range.
Possibly. Outputting and plotting the raw adc vaules in a graph will clearly show any saturation effects.
Starting with DC measurements is always a first start to see if your sampling frontend and ADC code does something that looks plausible. Circuits like this look deceptively simple, but all the parts have to be dimensioned right relative to each other for it to work well. Calculations on the ADC values hides all details of the sampling input. A good way to verify is to write a little oscilloscope function (including triggering, possibly on a low voltage channel only) Then verify whether your ADC frontend works by sampling some data both with a decent oscilloscope and with your gadget, and then compare the data. Probably easiest on your PC, but you may be able to upload the raw ADC values to your scope to a "reference channel". Also do this with badly distorted signals (high crest factor) and verify your ADC can capture the waveform properly.