perf: prefer recursion over multiple Keyword.has_key?

[katafrakt]

Less obvious (for me) than #128, but also seems like a quick perf win. Perhaps surprisingly, using tail recursion is generally faster than Keyword.has_key? by about 10%. In this case though the most important thing is that we iterate over the list only once, instead of 2-3 times.

The benchmark shows 7-8% improvement on parsing files.

[mhanberg]

Can you share the benchmarking method you are using? Could be useful for the repo.

[katafrakt]

I used existing bench.exs with some minor modifications and ran it before and after the changes. So not anything super-scientific, but few runs like that gave generally similar results.

For this PR I also created a standalone test with Benchee to see that recursive version is actually faster than Keyword.has_key? (although Benchee warns that for such fast operations it might not be super accurate).