Process Report in a separate thread

[lrntct]

• Make sure that SimulationData contains only data structures that can be deepcopied
• Remove the background processing of raster from Igis
• Add a new worker function that pull from the queue and process everything that is now done in step()
• In Report.__init__ create the queue and the thread
• Refactor Report.step() to make a deepcopy of SimulationData and last_step, and add them to the writing queue
• Decide how to handle the update of self.output_maplist and self.vector_drainage_maplist: either update them before adding values to the writing queue, at the risk of having a wrong list if something fails in the writing thread, or create a second queue where the worker will add the names of the written maps. The main thread will then take the values from the queue and update the two lists.
• In Report.end(), join the queue
• in __init__, add an option for the writing to be blocking (current implementation), or non-blocking

[lrntct]

Careful benchmarking would need to be done, as the competing thread could hurt performance.
Maybe with a new version of python without GIL could help.
Using process instead of thread is likely not appropriate because it would mean pickling the arrays to put them in the queue, and the inter-process queue has a size limit.