Idea of optimizing removal from index

[jschmieg]

There are 2 steps in removal from index:

1. findRecord(key) with Index Cursor - returns RecordID
2. unindex(key, RecordID) in SortedDataInterface, which also looks for key in tree.
   In case of many duplicated keys in tree, searching is very non-optimal, can be compared to looking in list (~O(n)).
   Idea is to cache location of (key,recordID) pair in IndexCursor : seek() or next() methods, and then reuse them in unindex() method, instead of looking for them again.
   Cache may be placed in RecoveryUnit instance created for current operation.