'pipeline' containers ready before 'remove()' algorithms

Change of plan: The development of a parallel 'remove' family of algorithm is delayed. Before that, two new container types called 'pipeline_circular' and 'pipeline_archiver' will be released. These data types are basically pipelines, fully orthogonal with STL containers and algorithms, and all the RPA model for multi-threading. As usual, they can adapt to any type of thread, mutex or condition variable. It makes stream processing orthogonal with STL parallel execution.

Releasing pipelines before 'remove' algorithm has some tactical advantages :
(1) This ensures that existing code is bug-free and flexible enough, (2) This allows more sophisticated tests (transform piped into remove piped into accumulate, each step running in several threads, etc...) (3) This allows to run plain sequential functors in pipeline as long as they have the required interface (aka stream processing)

This release should be ready in June (Delayed of two months). It includes too:
(1) More debugging, optimization, documentation (doxygen HTML doc, texinfo documents).
(2) The possibility to run detached algorithms (aka 'futures'): This feature was planned for a long-time by having a 'main thread' in the thread_tree datatype, but it was not used yet.