tests: extend benchmark with more parameters and better output

* run mutliple custom models via cli
* run multiple backends via cli
* print outputs as markdown table