While I could achieve all of that with OS driver and plover the sad reality is that I am unable to install my own software on my corporate PC and I would like to have consistent environment at my home and office.
Do keep in mind that /usr/share/dict/words is about 10x larger than the flash on microcontrollers. This is why most people do the steno conversion in software, not on the keyboard itself.
If it wasn't clear yet it is my personal project, I don't plan to make product from it.
The F103 is also interesting in that it is well-supported by Rust and Tinygo, if C isn't your thing. I picked it because QMK supports it (and works great), but I'm actually planning to write the firmware in Go and ditch all the things that bug me about QMK. However the USB stack is not there and I dread writing it. I might pull a QMK and just have ChibiOS run my Go code ;)
I was initially thinking about something that is already bluetooth enabled like ESP32 or CC2640R2 (I have development boards for both of these) but the design suddenly gets exponentially more complex if you want to get it right. Also, neither has a package that I can easily solder. I know it is just aesthetics, but I really like a clean board.
For final integration tests, you can use an additional microcontroller to be the key matrix. Instead of wiring up keyswitches, wire the microcontroller to the matrix pins, and have it accept commands over the serial port to set the matrix to a certain state. This way, you can do a full end-to-end test on your real hardware.
The only place USB gadget emulation might be helpful is in writing the code that switches the keyboard from compatibility mode (6KRO) to the NKRO mode. It's been done a million times, though, so there is plenty of library code to just copy.