Replace autocannon about benchmarks

[mcollina]

The focus of this is to measure the overhead introduced by the framework on top of Node.js. There is no other goal and the results are not meant to be measured in absolute terms (never do that).

We use autocannon because it's way easier to manipulate in js. The result are not absolute, but totally relevant in respect to all the other frameworks. As you can see, we benchmarks only frameworks that do not have any C++ parts.

We do not aim to be techempower.