Speed up findLongestMatch by avoiding a lot of allocations

[FrnchFrgg]

Instead of using maps that are recreated anew at each iteration of the loop, allocate two slices once and reuse the old one after cleaning it up. This allocates 35% less and uses 27% less time on real-world cases, and on synthetic benchmarks with long files to diff and a bigger number of similar lines one can observe one tenth the allocations and half the running time.
This patch series contain improvements to both the []strings and [][]byte implementations.