Instead of starting at the beginning, start at the end. Start with what is known or fixed and work from there. The arduino to use appears to be flexible.
It sounds like the final slaves are decided. How many can you get working at once? Is that enough? Probably not. So how can you connect a bunch of those “modules” (arduino-slave combos) do you need? How to connect them together to your Arduino Mega data concentrator becomes the next question. Preferably a method faster than I2C.
This could reduce the Arduino needed to a cheaper mini. Also the arrangement allows for multiple size systems using the same arduino-slave module and only having to reprogram or reconfigure the concentrator head.