Replace wrapperExecutionLock by finer grained locks about dmtcp

[karya0]

In principle, I agree with more fine grained locks and then some notion of lock ranks. However, I have some concerns about the proposed scheme(s). It could very well be that my understanding is a bit off :-).

What are some other examples of the ckpt-thread or other threads doing something sensitive? The ckpt-thread itself should never be doing malloc/free. If other threads use malloc, then they can use the alloc plugin that would take care of it.

I am not sure if creating a new state (ZOMBIE) alone would solve the exitInProgress problem. There is a possibility of a potential race between ckpt-thread setting state to RUNNING at the same time as the user thread setting it to ZOMBIE. The solution would then require us to first create a state diagram of available states and the rules about switching between them (e.g., you can go from ZOMBIE to RUNNING, etc.). This has been a long pending idea which we never go around to implement :-).