Moved AL into SeqDesign and restructured score combination

The functions to choose the next training sample in AL were moved from Engine to SeqDesign. The exploration and exploitation were made more distinct and their combination in the function SeqDesign.choose_next_sample generalized to follow the same structure for all combinations of methods. Voronoi was fully disabled as an exploration option. The tests were moved accordingly.

examples/analytical-function/example_analytical_function_testSequential.py

@@ -234,18 +234,28 @@ if __name__ == "__main__":
     This part of the code tests the available combinations of the 
     sequential strategies.
     The following combinations have remaining issues to be solved:
-        - exploitation = 'BayesOptDesign' has circular dependency with the engine class
         - exploration = 'Voronoi' creates mismatching numbers of candidates and weights
         - exploration = 'loocv' needs MetaModel.create_ModelError, which does not exist
         - exploration = 'dual annealing' restricted in what it allows as exploitation methods
         - tradeoff = 'adaptive' perhaps not possible in the first AL iteration?
         
     The following combinations are running through:
-        - BayesActDesign quite fast
-        - VarOptDesign as well
+        Tradeoff
+        - None
+        - equal
+        - epsilon-decreasing
+        Exploration
+        - random
+        - global_mc
+        Exploitation
+        - BayesActDesign
+        - VarOptDesign
+        - alphabetic
+        - space-filling
         
     Performance notes:
         - BayesOptDesign slow in the OptBayesianDesign iterations
+        - exploitation = 'BayesOptDesign' has circular dependency with the engine class
         
     # TODO: user-defined options were left out in this test for both 
     sampling-methods and exploitation schemes.