Both algorithms are included since the LTA appears to beat the SLTA for
all normal uses, but the SLTA beats the LTA for very large dense graphs.
Leave tools in the benchmark code to allow users to determine which one
they want to use for their data.
The initial test cases are from Robin Eklind's decomp project, the DOT
files were generated using his instructions:
git clone https://github.com/decomp/testdata
go get github.com/decomp/decomp/cmd/ll2dot
cd testdata/pathological/testdata
ll2dot *.ll
