Statically Building Backend(s) Into Llama.cpp App

[UsamaAshraf]

Seems that when running inference llama.cpp dynamically loads backend (what llama.cpp calls "backend") libraries (.so, .dll files) at runtime.

I'm using llama.cpp as a dependency for a MacOS Swift app, but I guess this question would apply to any app where we're trying to use the C++: can we statically embed the backend libraries into our executables?
E.g. the llama-simple executable, which we can build from the examples, would become self-contained and a bit bigger since the backend libraries would be compiled with it rather than attempting to link the libraries at runtime.

Also, shouldn't static linking of the backend libs be the sensible default? Unless we know for a fact we'll run into a high memory footprint which maybe mitigated by multiple apps sharing backend libs on the same machine instead of each having their own embedded backend libs.

This discussion somewhat touched on it.
#7631