--vec overhead from thread generation

[philipportner]

Currently, when running DAPHNE with the --vec flag, the WorkerCPU constructor creates a new std::thread for every call to the constructor.

Meaning, for every call to a _vectorizedPipeline__* we create new std::threads. For scripts that call have lots of vectorized pipelines, this induces a significant overhead in execution time when running with --vec.

Possible solutions:

• introduce a thread pool, this could be coupled with --vec overhead because of poor hwloc usage #877, maybe when creating the thread pool we once fetch the topology
• other solutions

[philipportner]

This is part of the problem that causes a script that runs in ~0.25s without --vec to explode up to ~6.2s with --vec.