these tasks have some shared data dependencies so they cannot be done
simultaneously. Future work should untangle these data dependencies so
that more can be done in parallel.

for now this commit ensures correctness by making linker input parsing
and codegen tasks part of the same queue.