Evaluate what you're actual foot-print is likely to be and pick a part with enough head-room to accommodate current and future code expectations, and mission creep.
Pick parts which have immediate families members in the same form and fit as the current part you buy, but with larger RAM/FLASH sizes. Often price stratification is achieved by selling the same die tested to different levels of capacity, ie die might have 512 KB, and options to get 128 and 256 KB versions, where time on tester is a significant contributor to throughput / cost.
Most products use firmwares that have optimization turned on, as this is most efficient. The compiler can fold the code, pull code that doesn't change out of loops, etc, and change ordering, use registers instead of memory to hold transient values. Without optimization you can debug code in a linear 1:1 relationship between what you wrote, and what the MCU executes. Lines of C code typically translate to multiple machine instructions, these can be a linear blob, or be mixed up somewhat for the sake of efficiency, the former is a lot easier for most humans to visualize.
