Well, it does work doesn't it? It is somewhat difficult to understand the order the scripts operate in.
For example, the serial ports can be opened in a way that a second program cannot open them. Furthermore it should only be probing ports that are suggested as possible gps based on udev rules.
Well it was difficult to build scipy but perhaps tinycore 13 is easier? Did you have to build gfortran manually as well?
Once you shrink the filesystem then partition, be sure to write the rest of the disk with zero. like dd if=/dev/zero of=/zero
until it fills the disk, then delete this file. It helps with compressing the image. I prefer to use xz compression.
Anyway, I may move to your image if it really is all working, but I would not suggest it as "stable" for some time due to the major changes in packages. The current image is built on a system already proven over many miles. I have produced (perhaps final) of tinycore 9:
https://pypilot.org/images/tinypilot_2021_11_16.img.xz
The image was mostly for pi zero before which did not run opencpn well. Now it might be nice to provide opencpn via tinycore rather than a serial console (full of strange messages)
Will be excited to test the tinycore 13 with the pi zero 2 as well.