Add function to compute the mask at the split level and fix Ref #45

The projection from the left subtree onto right subtree on the split nodes is now correctly masked.

This is a fix for the problem in Ref. #45 (closed). To cover problems of higher dimension, the maximum dimension in the test fixture is increased to 5. This is a compromise because for dimension higher than that the test will take too long to run. However, separately, the fix is tested against problem with dimension up to 10 and degree up to 6 before hitting the memory ceiling (at least in my PC).