Code Optimization + Parallelization

[giovannidispoto]

This PR contains modifications that significantly affect performance, in particular:

• Parallelization of trajectory collection
• Correction of the tree dimension for the GAPS algorithm (avoids using a large)

Note: I currently have a doubt regarding the correct tree dimension for the p_gaps algorithm.

Minor:

Updated README.md and requirements.txt

[giovannidispoto]

In the last commit, I vectorized the split computation in order to speed up the algorithm. On my machine, I measured this improvement:

Old split time: 0.16466116905212402 seconds 
New split time: 0.0018467903137207031 seconds
Old split time: 0.15724492073059082 seconds 
New split time: 0.0017821788787841797 seconds
Old split time: 0.07813882827758789 seconds 
New split time: 0.0017809867858886719 seconds

I also added a simple test to verify that the results obtained with the old and new versions are identical (please use it without adam learning rate to avoid an error).