Ns/chore/fix zk thrpt elements

[nsarlin-zama]

PR content/description

Since the zk verification does not perform any pbs, the usual heuristic for throuhput elements count does not work. We use a different heuristic:

• zk verify is pooled (see refactor(zk): run pke_v2 verification inside dedicated thread pools #2310). On hpc7 we have 6 pools.
• We "fill" the pools with 64 elements, which experimentally gives a good throughput.

For verify_and_expand, we use the classical pbs count and we take the lowest value between both heuristics.

I also removed the zkv1 integer benches. This reduces a lot the bench time, allows to reduce clutter in the grafana dashboard and we bench zkv1 in the tfhe-zk-pok crates anyways