I've been benchmarking whole calls to eg `git annex whereis`, and that's
not ideal because git-annex has some startup overhead that I'm not
interested in benchmarking (right now), and often that overhead swamped
the things I wanted to benchmark, making it difficult to trust my results.

So, I've built a `git annex benchmark` command, that can benchmark any other
git-annex commands, without starting a new git-annex process. It uses
criterion to get statistically meaningful benchmark results. And operations
as fast as 10 ms can be benchmarked now, without needing to write any
special purpose benchmark code.

New results for this weeks's optimisations:

whereis on 1000 files........... 5% speedup  
whereis on 1 file............... 14% speedup  
info on dir with 1000 files..... 4% speedup  
local get ; drop of 1000 files.. 3% speedup