Faster builds

[mariusandra]

Makes builds 2x faster thanks to parallelization (except for the original single core Zero W)

• Parallelize build on device. This takes a full build on a Pi5 from 90sec to 50sec. At 3.5min for Zero W2, down from 8min.
• Cache parallelized build. This takes it down to 15sec on a Pi5. Under 50sec for Zero W2, down from 90 sec

We could probably save even more with cross compilation, but with under minute compilation and deployment times, it doesn't really matter anymore, at least at a small scale.

The nimc.Makefile is ready to be compiled inside dockcross, but getting that working inside a Dockerfile and/or or on all platforms sounds like a chore. I'd either have to make the image much larger and include build tools for all targets, or put docker inside docker. Then, to link to system libraries like lgpio or evdev, I also need to either know their exact versions, or scp the .a files from the client device onto the build server.

This complexity adds up, and since the alternative is to "wait a minute" (instead of waiting half a minute), I'll come back to it later. Perhaps with a stand alone build server.