cubicweb-worker #4200106 consider using subprocesses (through the subprocess cube) [validation pending]

As of today we can deploy "worker instances" and get our tasks picked up by these distributed instances, which is neat.
BUT for use on the same machine it would be desirable for any single worker instance to be able to use all available cores efficiently (which is impossible currently, because of the Python GIL).

Using subprocesses (possibly by default) would enable having as many "workers" as cores in a system without multiplying local instances.

prioritynormal
typetask
done in3.1.0
load1.000
load left0.000
closed by<not specified>
patch[views] give a better-than-default primary view for task and subprocesses [applied][subprocess] allow tasks to be run from within a subprocess [applied][script] add a script that will launch a task from a subprocess [applied][pkg] add the subprocess cube as a dependency [applied][api] switch to cubicweb 3.19 API (session -> cnx) [applied]