I have to admit when I looked at a picture of the trinket M0 on the Adafruit site I though it had more available pins than it does. Getting LMiC going is going to be tricky because you don’t really have I/O’s left to monitor the DIO pins typically used for detecting the end of transmission and the success of reception. It is possible to do this by polling, it would require slightly longer receive windows in compensation but that doesn’t cost as much power as one might think. The more practical problem is that this would require some code changes. And you don’t have much left for reading sensors (though you could use the LED pin as another chip select and read SPI things).
They’re really not more expensive in actuality. If you compare something like the ubiquitous STM32F103 “bluepill” to various nano/micro Arduino’s the cost is about the same, but the capability is quite different.
Mostly it’s a matter of inertia - though the point about that to note is that the Arduino-realm inertia of TTN and other LoRaWAN efforts is not and never really has been with ATmega328’s running LoRaWAN stacks. People have either used more capable MCUs also supportable by Arduino, or they’ve used a module with a built in LoRaWAN stack and had an ATmega only to collect and prepare data.