call() does not preserve float registers

[JayFoxRox]

If you intend to use floats in your calls, you have to manually clean-up and restore.

It has not been decided what the final calling convention for our call() is.
For now, assume that you always have to clean up the FPU, and this is a documentation issue.

There sould be a comment in the code, or some docs which make this clear. The GitHub issue is a temporary measure.
Alternatively, if you feel strongly about this, you can implement float cleanup. But I personally believe it would overcomplicate an already complicated assembler function.