Improve performance of descendant determination

[paulolaup]

IS:

The current performance of the descendant counting algorithm is not acceptable for large value sets based on code systems with complex hierarchies (e.g. SNOMED CT). The issue seems to be that results already calculated during determination of the descendant count of a previous concepts are not used for other concepts. The result is that there are many duplicate calculations causing a large performance hit for large hierarchies.

TODO:

Implement lookup for already calculated counts and use them to calculate counts of ancestor concepts.