Speed up Python import of small models

[dweindl]

Import time for small models is determined mostly by compiling and especially linking. Pretty sure there is room for improvement...

Potential starting points:

• [ ] reduce swig interface
• [ ] limiting symbol visibility

[FFroehlich]

From what I can tell it's primarily the compilation of individual model files. I think we can probably achieve more through

• [ ] optimizer header linkage (
• [ ] use ccache when possible
• [ ] custom/reduced compiler/linker flags specific to model compilation