(Depends on the Z3 API PR (not because there are real functional dependencies but simply because I merged the two at some point to be able to test both together))

Building on Malte's earlier attempt to get parallel branch verification, this PR adds a command line parameter --parallelizeBranches that directs Silicon to verify different branches in parallel.
We use a ForkJoinPool for this which uses a work stealing algorithm, so every worker has its own Z3 instance and its own work queue. When it reaches a branch, it puts the else-branch into its own queue s.t. other idle workers can take it; if noone does, it will take the else-branch-task from the beginning of its own queue and verify it itself.
Workers that become idle will steal from the end of the queue of another worker and therefore get the oldest task that worker still has left over, which tends to be the biggest task in the queue, which should avoid switching back and forth between workers for tiny tasks.
Every task switch requires updating the worker's solver to the branch's current state, i.e., popping all its current path conditions and pushing all path conditions from the task it's taking over.

Speedups obviously depend on the program and the machine, but if you have a bunch of cores lying around and your program has lots of branching, things will become MUCH faster (e.g. 6x for Nagini VerifyThis examples on my desktop at home).