The way @fptw was handled has been wrong for a long time but I finally figured out. This updates all the backends to do the right thing.

Other more minor changes:

• The code that reads the CPU state has been updated to be compatible with snapshot. It also should be now able to load precise values for 80-bit registers. Loading 'old' dumps should also work as they were before (with the buggy code) to not break them
• Updated bxcpu to be able to load @fpst registers with a frac / exponent (cf yrp604/bochscpu@cab8051)
• Minor updates to the README
• Update the CI to get rid of warnings as well as removing the CodeQL scanning (it hasn't reported a single hit since it's been turned on -_-)