Merge branch 'vlasta-candidates-leaf-hulls-static-and-candidates' of https://gitlab.fi.muni.cz/disa/public/hulls/mh-tree into vlasta-candidates-leaf-hulls-static-and-candidates

* Fixed computation of hulls during node split (duplicate objects in lists)

Multiway insertion algorithm (provided by the cmd-line argument --multiway YES; or by an operation argument "multiway" (boolean))
* Use candidate objects in internal nodes in the coverage test
* Method of updating hulls after each split added -- it recomputes hulls from the scratch using all data objects rooted from the node that has been split or a child has been added into)
* Fancy print of whole tree structure added

* New InsertType GREEDY_AFTER_SPLIT - it recalculates hulls after each node split
* Code clean up

* Changed management of candidates in internal nodes: Node.insertCandidate - replace the closest candidate with the new (better (farther)) candidate
* Changed default of recomputeDuringSplit to false -- if true, it checks and changes candidates among siblings in Node.getRelevantCandidateHullObjects
* Change: When a node is split and the new nodes fit into the parent, the parent's hull is updated with candidate objects.
* Change: internal node's isCovered test takes into account the candidates in a way that an object non-covered by hull is taken as covered if its rank is smaller than at least one candidate of the node.

* Added cmd args for recompute during split/insert
* Minor code clean-up

* Minor updates (COVER_DIST added to list of supported functions and fat factor computation reports the average number of accessed objects per exact match query)

* Fat-factor extended with the average number of accessed objects to get an exact match
* Code clean-up in InternalNode -- calling the custom object-to-node ranking function
* NEW: COVER_DIST object-to-node distance based on isCovered definition

* Fixes of MHTree to accept zero objects in bulk loading.
* Optimization in fat-factor computation

* leaf hulls are updated on insertion but rather any non-covered object is added to the leaf candidates. isCovered test then respect it.

* Bug fix in approxKNN search operation
* Fix in creating a hull not with the required minimum of hull objects.
* Serialize a populated index to disk has been added

* Code clean up

* Bug fixes: ?
** Parameter --recompute is no used (it was ignored and always set to true)
** MHTree.recomputeAncestors is not used in test to update parent hulls in balancedInsert but rather recomputeCandidatesDuringSplit is tested

* A non-covered object is added to candidates of one child only!
* Updating hulls only when nodes are split

Added ORIGINAL distance function that calls rank by David Prochazka and its use in the MH-tree code by Josef Podany

* RunBenchmark fixed to use all CLI arguments and some added (query file, e.g.)
* SIMPLE_SPLIT added to Preprocessor
* Fixed mess in naming InsertType
* Code refactoring, comments added

This fixes the redistribution of candidate
hull objects during node splitting.

* Fixed bug causing SUM_DIST to always return -0.5.
* Extended the ObjectToNodeDistance class to accept
  also a list of objects instead of node.

Same ranking function is used for inserting
objects and searching in the MH-Tree.