Cleaner ADC measurements which of these ways is better

A few ideas:

A LC filter (inductor and capacitor) for AVCC is very good, but not always needed.
Do not use a RC filter for AVCC.

It is not clear what is connected to AREF. I hope you did not connect the 3.4 V of VCC to it :confused:
Keep AREF open, but do add a capacitor of 100nF between AREF and GND. That does help.

I like the idea by @ted to change the switching regulator to 5V and add a LDO for 3.3V.

You could use the internal voltage reference of 1.1V. It is in the datasheet how much it rejects the VCC ripple. I forgot what it was, but it was a good number. Don't use the internal voltage reference when using an analog ratiometric sensor.

Use many samples to take the average if you have enough time for that.
When measuring a battery voltage or the light intensity or something else that is slow, the average of only 5 samples reduces the noise a lot. Sometimes I use the average of 10000 samples, but that's nuts. About 5 to 30 samples is often enough.

The best option depends on the sensor that you want to read. Reading a ratiometric sensor is not the same as reading a battery voltage.