Schematics for IMT ProjectC is more accurate.
I hope not because you have shorted all the sensors together so you can never distinguish between sensors. Diagram B is better although very badly drawn.
I though having the 1M resistor in paralell to the entire multiplexing process would be ok.
It is not because it leaves all the other inputs floating and is only applied to the one sensor when it is switched through for reading.
Not only does it minimise cross talk it also limits the voltage into the multiplexer to a safe level.
When ever you get high impedance inputs being coupled together inductively ( due to close proximity ) the transformer effect will induce voltages in the other lines. This is cross talk.
Thinking about your previous observations with the reading of a floating input.
The readings you showed I think were a result of a read, print cycle. The print statement takes time and so the readings you were getting were not very fast. Make an array and read, save in array and delay for 1mS. Then when the array is filled up ( 50 samples should do) print them out. You will see a very different signal, one much more like a sin wave at 50 or 60Hz depending on your mains frequency.
For a further illustration get an oscilloscope and set it on 1V per mm. Then touch the end of the probe. You will see a huge wave, that is you acting as an antenna picking up the noise.