This increases code size by 1 instruction (2 bytes) because LLVM isn't
yet smart enough to recognize that it doesn't need to clear a register
to use 0: it can just use r1 which is always 0 according to the
convention. It makes initialization a lot easier to read, however.
This allows the use of some compiler-generated builtins that are
hopefully compatible with LLVM. Example: println(uint8(foo))
Code size is unchanged normally but of course compiler builtins will
increase code size when actually used (for example with division).
This is kind of dirty with that huge list of linker params, but it works
and it produces smaller object files (probably because GCC is better
optimized for size).
