No process spanning for n_jobs == 1 + more robust worker termination

1. We now do not span a new process when the number of workers is just one. This enables effective debugging for the code that is executed through mtasklite. In contrast, when a new process is started, setting debugger breakpoints and debugging that separate process works poorly (often not all).
2. We implemented a more robust mechanism for emergency exit through an additional control queue.

Fixing default chunk sizes (greatly improved performance).

If the chunk size is not specified, using the default size of one leads to no parallelization in most cases. The queue size should be >= the number of workers and it can help to "prefill" it with extra elements.

Fixes

1. Fixing the bugs (not immediate shutdown in the case of exceptions, and not updating the bar properly) for unbounded mode.
2. Deprecating task timeouts.

v0.2.1

This release fixes several documentation issues and removes a debug statement from a unittest module.