Draft: Restrict packet-selecting logic test to 32-bit scalars

@rmlarsen1 This fixes the compilation issue.

However, there's a runtime crash in vectorization_logic_1:

 Expected Traversal == InnerVectorizedTraversal got LinearVectorizedTraversal
 Expected Unrolling == CompleteUnrolling got CompleteUnrolling
Test vectorization_logic<std::complex<double> >::run() failed in /home/bitonic/src/eigen/test/vectorization_logic.cpp (181)
    test_assign(Vector3(),Vector3()+Vector3(), sizeof(Scalar)==16 ? InnerVectorizedTraversal : (1 ? LinearVectorizedTraversal : LinearTraversal), CompleteUnrolling)
Stack:
  - vectorization_logic<std::complex<double> >::run()
  - vectorization_logic
  - Seed: 1638309762

It's not unsurprising that something in this area changes -- we might now select full packets and have leftovers, which would select LinearVectorizedTraversal.

I now need to go, but I can probably go through those tomorrow. The problem is that it's very hard to make changes in that test harness with confidence if I don't run all the interesting configurations. Is there an easy way to do so?

Francesco