Back-end fine-grained benchmarks

[elefthei]

In bench.rs there is a microbenchmark called prove_verify this is very corse grained and also doesn't seem to work.
Can we make several benchmarks for

1. Initialization
2. Proving
3. Verifying

That do not take solving into account (we already benchmark this correctly).

[elefthei]

@jkwoods this is probably very easy to do at this point? :)