Simplify, fix and improve similar images algorithm

[qarmin]

• New algorithm of finding similar images - should be faster, give a little better results and fix some problems, it is also easier to understand
• Fixes problem with crashing when using reference folders, introduced in recent gui cleanups
• Added tests
• App should find more broken pdf files

Times in mm:ss format

78000 tested image files, hash size 16 - Nearest - 40 - double gradient:

Old algorithm - 8:13  
Current implementation - 4:42
First implementation of current algorithm when computations worked mostly in one thread  - 16:51

78000 tested image files, hash size 16 - Nearest - 10 - double gradient:

Old algorithm - 0:52
Current implementation - 0:35

Initial implementation:

• Calculate hashes
• Find similar hashes for each hash(one threaded)
• If hash was not used before, use it now

Previous implementation:

• Calculate hashes
• Split hashes to check into n + 1 parts(where n is number of threads)
• Check this chunks in parallel
• Maintain for each thread list of hashes are used as "originals" and info about similarity between files
• Compare one hash with all other (multithreaded)
• Fill and compare results from previous comparisons (multithreaded)
• Connect all results from initial chunks and compare results to each other (one threaded)

Current implementation:

• Calculate hashes
• Split hashes into groups of 1000 items
• Maintain only one list of hashes used as "originals" and
• Compare each hash from this group to every other (multithreaded)
• Throw out results that in 100% will not be used(e.g. hashes with too small similarity) (multithreaded)
• Connect results from 1000 items with previous results (one threaded)