Worth still considering @davek79 excellent suggestion of keeping the comms as a slave device, and the ESP32 as a comms coprocessor? Sensible to keep safety critical on a device that isn’t busy talking to others.
Oddly, the community picture of an STM32Fxxx based flight controller is a brilliant starting point. ST micros are well used (and solid), Atollic give a free and brilliant IDE, and the debugger is cheap. Temperature is 105C maximum, but that is a trustable 105C… Also, you get a 6/9DOF plus a barometer interface on most similar boards (So you can crib code from existing - and well written - resources like Betaflight). The 6DOF is the killer feature for me over a standard paramotor, as it has the ability to thrust vector your motors in conditions like undetected motor failure twist, launch where the pilot face plants (a game changer to have a safety kill when the pilot falls face first on to his throttle) etc. You can quickly make gasoline machines look dated and dangerous.
For all of your failure cases, please be careful with all of the inter board interfaces. Some diagrams show i2c etc, which are great, but they MUST have protection for short circuits,ESD (aircraft props generate a LOT of static!) Etc. Please consider using isolated CAN or RS485. I’m not preaching, just do an analysis where you do a short circuit supply to ground on the hand throttle (damaged cable), and an open circuit ground to one of the ESCs. You may find that you end up with some voltages at IO pins that will cause considerable current flow and a loss of the whole paramotor on inputs from, for example, the tacho signal from the ESC, or outputs to the ESC. A normal controller would have a tracking regulator feeding the hand throttle to protect the FC from loss of power in the short circuit case, and the ability to check the hand throttle for failures (supply overcurrent/voltage out of range). Similarly, a digital input from another board may well have a low cost HC14 based inverting schottky buffer and anti-alias filter, plus sufficient series resistance and double diode protection such that a 48V short on an input would cause no stress or issues.
All the best with it, looking forward to seeing progress.