feat(metric): add Quadtree decomposition

[kargaranamir]

• metric
• support types for all variables
• doc
• provide tests
• needs to check some test cases vs python2
• add comments for functions
• register metric in metrics.json
• run evaluation

[kargaranamir]

This pull request refactors the Quadtree decomposition metric in the AIM legacy based on this paper. In all test cases, the new implementation yields the same results as the legacy implementation. I think the metric is ready to review.

• I add another different way (it's older) of computing the Quadtree decomposition in the metric besides the legacy one. However, all the metrics are computed with the quadtree decomposition function in legacy since the paper used it as well.

  • the legacy metric used the Shanon entropy over the colored image (same as the paper). (duckduckgo example).
  • the new way I added is an old one, just using the original way of constructing quadtree based on computing std over the gray image. (most implementations on the web in different applications follow the original way, so I thought it's good to keep it). (duckduckgo example).

• Bugs I found:

  • R_j and T_j variables are swapped in the symmetry function.

• Issues:

  • Thresholds heavily influenced by websites as legacy code says or based on nothing!
  • Equilibrium gives high results as the legacy code says as well. see Histogram.
  • I think the symmetry results are not promising based on the histogram.