Additional benchmarks

[ksahlin]

In order not to overfit for future parameter optimisation efforts (e.g., #272), it would be good to include:

1. A benchmark on a metagenomic dataset with several similar strains.
2. An analysis that focuses on accuracy around only around variant sites, similar to this benchmark (or we use the linked benchmark as is).

(Not intended for @marcelm as you have enough on your plate, I will try to find time to set this up myself or find someone else)