Explore adding mode where tasks run sequentially or block

[msimberg]

As a debugging aid, it may be useful to have a mode where all tasks either are forced to run on a single thread or where each supposedly asynchronous task launch is forced to be blocking, similar to CUDA_LAUNCH_BLOCKING. It's not clear yet exactly how useful such a mode would be, so this would need more motivational examples before implementing, but I'm putting the idea out there for now.