Optimize/rewrite llvmir2hll's CopyPropagation phase to be faster

Currently, `CopyPropagation` from `llvm2hll` is the biggest bottleneck - it often takes huge amount of time and memory (#13) to finish. It needs to be either optimized, or rewritten, in order to run faster.

Start by locating the CopyPropagation phase in llvm2hll and profiling its runtime and memory use, including the behavior described in #13. Determine whether the existing phase can be optimized or needs rewriting; done means it completes faster and uses less memory on representative inputs, with comparable propagation results.

Written by the indexing model from the issue text.