I’d echo the comment that it would be a great learning exercise, but I’m not sure if you would end up saving that much money in the long run.
I very much look at it as a cost comparison. Which is more costly, spending the time and revisions to get the discrete+MCU solution working, or using an all-in-one external IC? If you’re building 10k or 1M quantities, then definitely spend the time to save every penny. However, if you are building 100 or less, it might be overall cost effective to just use the integrated solution.
One other consideration: When I do designs, especially in areas where I don’t have recent and immediately relevant experience, I will always choose ICs like the L6208 or the Trinamic drivers that Phil mentioned. My rationale being that I would rather have fewer “degrees of freedom” when implementing something for the first time because it gives me a much higher likelihood of the design working on the first or second revision. What I mean by degrees of freedom is not the IMU sense, but in the design sense. By choosing discrete MOSFETs to drive the windings, and by implementing the waveforms and timings yourself, there are that many more things that can go wrong. (The same rationale applies to development boards, either made by the manufacturer or by companies like Adafruit/Sparkfun/etc.)