customized fit output

This MR decouples the output buffer (receives fit results from pymca) from the output handler (saves results in Nexus or to a queue).

See outputbuffer_context how buffer and handler work together.

When use a queue handler, you can redirect the results to another handler (e.g. Nexus) with consume_handler_queue.

I parallelized the MultiDetectorFit class with the buffer/handler decoupling. It uses a queue handler in each subprocess and one Nexus handler in the main process.

I also expanded test_single_detector_fit to test different ways of fitting.