VQE should batch per default

[Cryoris]

What should we add?

The VQE is most commonly used with SPSA or gradient descent, which require 2 or (up to) 2 * num_parameters circuit evaluations per iterations, respectively. Since the new execution model is shifting from using the runtime VQE program to the primitive-based VQE in Terra, we should also enable batching as many circuits per default as possible. Otherwise the standard usage of the VQE will be much slower than necessary (~2 or 2 * num_parameters to slow 😅).

This is something the runtime VQE also did, maybe we can use the same defaults?

[Cryoris]

Actually, maybe we can consider this as performance bug and put it in the 0.22.1 release for November 🤔

[mtreinish]

As long as we don't add any new flags or change the api I think it's fine to backport and include this in 0.22.1 as a performance fix.