adjust recursive calculation using standard gates

[ed255]

I realized that the measure of verification cost was done on a circuit with just the NoopGate, but every time we add a gate to the set the verifier needs to perform the corresponding evaluation, so more gates add a constant overhead.
Now the measure is done with the same set of gates we use to define the recursively verifiable circuit.
With this we get a much better estimation and reduce the accuracy error of the linear regression to 2% (from 8%)

The overhead between verifying any proof with just the NoopGate and verifying a proof with all these standard gates is about 400 num_gates (rows), no matter the circuit size.

This needs to be updated once we add the gates from #236

Blocked by #236

I've extended the PR to:
resolve #253

[arnaucube]

LGTM, will need to check if we would need to add a way to define extra custom gates for cases like the ecdsa introduction gadget

[ed255]

LGTM, will need to check if we would need to add a way to define extra custom gates for cases like the ecdsa introduction gadget

We definitely need to add the custom gates defined in #236 to the list and redo the linear regression to get accurate values. For this reason I'll leave this PR unmerged until #236 is merged so that I can adjust it :)