Stop. If you have many cells, you don't use many individual voltage dividers.

If you have a submodule handling 8 cells, then you can design with 25V and after the mux you have a single 10:1 divider to get into 2.5V range of your ADC. So the components of that voltage divider are common for all 8 cells. And that divider, the quality of the voltage reference and the linearity and stair errors of the ADC are what affects the quality you get. But you handle the next 8 cells with a different ground, so you don't accumulate any errors between the modules.

The only thing you need to think about within a submodule is that you don't measure all voltages concurrently, so you may have to run through your 8 cells multiple times and then average to decide the cell voltages - the reason is that you may have load changes while you measure and within a submodule, the cell voltage of the 7th cell is the voltage over 7 cells - the voltage over 6 cells.

The special case with a flying capacitor means that each submodule is suddenly just once cell large, and in that case you do measure the "real" cell voltage for the specific load the cell had when the capacitor was charging. The LT chips do perform similarly - they move both measurement probes when measuring the individual cells in the subchain, which allows them to take advantage of the full ADC resolution instead of splitting the ADC resolution over the number of cells in the subchains.

If you want to, you can take advantage of the low price of microcontrollers with 10-bit ADC and use many processors - each operating at a different ground reference and being powered by that specific subset of the battery cells. So you can have one processor, one voltage reference (if probably best with a processor with internal band-gap reference), one voltage regulator and one 8-to-1 MUX, 1 opto-isolator for each set of 8 cells. The total cost would still be very, very low for a larger vehicle running with 200+ lead-acid cells.